Computers doi: 10.3390/computers15070457

Authors: Yuhang Zhang Yiting Zhao Yujing Meng Jingqi Li Tianze Zhang Ying Zhang

Electricity load forecasting is essential for balancing energy supply and demand, reducing energy waste, and maintaining power grid stability. Accurate forecasts enable power utilities to optimize energy dispatch and mitigate the risk of supply shortages or outages. However, conventional forecasting methods often struggle to capture highly nonlinear local fluctuations in electricity consumption and long-term temporal dependencies. To address these challenges, this study proposes MSCNN-ResLSTM, a hybrid model for multi-step electricity load forecasting. The proposed model integrates Multi-Scale Convolutional Neural Networks (MSCNNs) to extract local time-series features at multiple temporal scales, residual blocks (ResBlocks) to enhance feature representation through residual connections, and Long Short-Term Memory (LSTM) networks to model long-range temporal dependencies. To comprehensively evaluate its effectiveness, a cross-paradigm experimental framework is established in which MSCNN-ResLSTM is compared with seven representative benchmark models from three methodological categories: traditional machine learning (Extreme Gradient Boosting-XGBoost), classical recurrent and convolutional neural networks (LSTM, Temporal Convolutional Network-TCN, CNN-LSTM, MSCNN-LSTM, and Direct LSTM (Seq2Seq)), and self-attention-based architectures (Transformer). Experimental results show that MSCNN-ResLSTM achieves higher forecasting accuracy and greater stability across the full 24-step prediction horizon, consistently outperforming all competing baselines while effectively suppressing recursive error propagation.

Computers doi: 10.3390/computers15070456

Authors: Adam Koty Abbass Ahmat Habiba Chaoui

This paper presents a systematic literature review examining how blockchain technologies can enhance the security and performance of multi-cloud systems. Multi-cloud architectures offer resilience, scalability, and flexibility; however, they also pose complex security challenges related to APIs, service-level agreements (SLAs), orchestration, and authentication. The promise of blockchain technology to improve the security and transparency of numerous applications, including cloud storage systems, has attracted considerable attention in recent years. Much research has focused on decentralized storage in cloud environments, spanning supply chains, FinTech, healthcare, and education. Still, the integration of blockchain with the cloud and its potential to enhance security and performance warrant an in-depth study. Using the PRISMA methodology, a structured search was conducted across six major scientific databases, including IEEE, ACM Digital Library, ScienceDirect, Scopus, Web of Science, and IJIMAI. Twenty-four primary papers published between 2019 and 2025 were selected for analysis after clear inclusion and exclusion criteria were applied. This review examines the security dimensions in multi-cloud environments—architectural vulnerabilities, API security, authentication, orchestration and automation vulnerabilities, SLAs, and cybersecurity compliance issues—in relation to blockchain technology. Based on the identified gaps, we propose a SOC-LLM-augmented security architecture that integrates blockchain-based evidence integrity, statistical anomaly detection, machine learning, large language models, and autonomous AI agents to enable intelligent attack detection and response. The proposed framework introduces specialized agents for detection, correlation, threat intelligence retrieval, blockchain evidence validation, explanation generation, and response planning. The analysis shows that integrating SOC-LLM capabilities with blockchain can move multi-cloud security from passive auditability toward proactive, explainable, and human-in-the-loop cyber defense. Finally, this paper discusses open challenges, including LLM hallucination, data scarcity, real-time scalability, evaluation standardization, and trustworthy deployment in critical multi-cloud infrastructures. The study’s conclusion highlights research gaps and suggests future lines of inquiry concerning scalable blockchain architectures and the incorporation of AI for proactive cloud security monitoring.

Computers doi: 10.3390/computers15070455

Authors: Ahmed Lateef Salih Al-Karawi Rafet Akdeniz

Service-Based Architectures (SBAs) in 5G core and cloud-native deployments require differentiated treatment for service classes with heterogeneous latency, reliability, and throughput expectations. Although HTTP/3 over QUIC is an important evolution of the HTTP ecosystem, HTTP/2 remains operationally relevant in SBA environments where TCP/TLS-based infrastructures and 3GPP service-based interfaces continue to rely on HTTP/2 communication. This paper therefore focuses on HTTP/2 priority signaling and the problem of translating application-level Service Level Agreement (SLA) policies into protocol-level priority metadata. To address this problem, the paper presents an SLA-aware priority management system built around the RFC 9218 extensible prioritization scheme, specifically its urgency and incremental parameters. The system integrates three coordinated subsystems: a rule-based Priority Classification Engine (PCE), a feedback-driven Dynamic Priority Mapping Algorithm (DPMA), and a runtime priority-update manager that applies bounded priority adjustments under changing network and load conditions. The revised evaluation reports a 7200-observation baseline campaign covering four operating modes, ten service classes, nine network profiles, and twenty repetitions per service–profile–mode combination, together with a 14,880-observation scalability and overhead campaign across increasing concurrent-stream levels. Compared with the unmanaged HTTP/2 baseline, DPMA reduced mean latency by 24.8%, P95 latency by 35.1%, P99 latency by 38.0%, and SLA violations by 19.9 percentage points. Compared with the legacy RFC 7540 baseline, DPMA reduced mean latency by 39.0%, P95 latency by 49.3%, P99 latency by 49.9%, and SLA violations by 21.1 percentage points. Compared with the static RFC 9218 baseline, DPMA reduced mean latency by 38.7%, P95 latency by 48.1%, P99 latency by 50.6%, and SLA violations by 21.4 percentage points. The scalability analysis shows that DPMA maintained P95 latency between 126.8 ms and 128.2 ms over the tested 1–100 concurrent-stream range, with priority-update decision overhead below 0.004 ms per request. The results indicate that SLA-aware use of RFC 9218 priority metadata can improve latency and SLA-compliance behavior in controlled SBA-like HTTP/2 environments while preserving a transparent and auditable prioritization policy.

Computers doi: 10.3390/computers15070454

Authors: Shuang Liu Huifeng Guo Rongyu Han Yajing Pang Gang Liu

Accurate emotion recognition is crucial for enhancing human–computer interaction, and brain–computer interface (BCI) technology offers an efficient means for emotion detection using the EEG signal. However, existing methods face significant challenges due to the inherent inter-individual differences and temporal variability of EEG data. To address these limitations, this paper introduces a multi-source domain adaptive algorithm based on dendrite net (DD-MSDA). The proposed model employs the dendrite network as a shared feature extractor to align feature distributions across multiple source domains, thereby capturing common features among diverse datasets. Experimental validation on cross-subject and cross-session tasks using the SEED and SEED-IV datasets demonstrates that DD-MSDA achieves highly competitive performance, outperforming all compared single-modal EEG-based domain adaptation methods. Moreover, the algorithm demonstrates statistically significant advantages over existing domain adaptation baselines in cross-dataset settings. These results highlight the consistent competitiveness of DD-MSDA across various cross-domain scenarios, and its unsupervised nature underscores its potential for practical online EEG emotion recognition applications.

Computers doi: 10.3390/computers15070453

Authors: Luís Nogueira Alice Resende

Cloud-native software-delivery pipelines increasingly rely on Policy-as-Code (PaC) to automate security, compliance, and governance enforcement. Although Policy-as-Code is widely adopted within Continuous Integration (CI) pipelines and Kubernetes admission-control frameworks, governance requirements are often implemented independently, potentially increasing maintenance effort and creating opportunities for policy drift. Despite the growing adoption of Policy-as-Code, comparatively little empirical evidence exists regarding the reuse of a shared policy-definition layer across complementary enforcement stages within the software-delivery lifecycle. This paper presents and empirically evaluates a reusable multi-stage Policy-as-Code enforcement model based on a shared policy-definition layer implemented using the Open Policy Agent (OPA) framework and its Rego policy language. Rather than proposing a new Policy-as-Code technology, the study investigates whether a shared policy-definition layer can support consistent policy enforcement across Continuous Integration validation and Kubernetes admission control. The model was evaluated using Conftest and OPA Gatekeeper through a structured experimental study comprising 29 Kubernetes manifests, 37 experimental scenarios, eight Kubernetes resource types, and 261 policy assertions covering representative cloud-native workload-governance requirements. Within the evaluated dataset, all intentionally introduced insecure configurations were correctly identified without observed false positives or false negatives. The shared policy-definition layer was successfully reused across both validation stages, while Kubernetes admission control mitigated all evaluated CI bypass scenarios by providing an independent deployment-time enforcement boundary. The results demonstrate that a shared policy-definition layer can support consistent policy enforcement across complementary enforcement stages while enabling policy reuse without requiring duplicate policy implementations within the evaluated environment. More broadly, the study contributes empirical evidence supporting policy reuse as a governance strategy for cloud-native software delivery and provides a reproducible foundation for future investigations involving larger datasets, broader governance-policy portfolios, alternative Policy-as-Code ecosystems, and production-scale deployments.

Computers doi: 10.3390/computers15070452

Authors: Shervin Zakeri Dimitri Konstantas Prasenjit Chatterjee

This study introduces the Painted Wolf Decision Optimizer (PWO), the first deterministic, bio-inspired decision framework for discrete multi-criteria decision making (MCDM) derived from specific observed decision behaviors of African wild dogs, including quorum sensing, dominance hierarchy, collective voting, and experience-based learning. Unlike conventional nature-inspired metaheuristics that rely on stochastic search across continuous domains, PWO defines a new class of Discrete Bio-Inspired Decision Operators. It formalizes key ethological mechanisms of Lycaon pictus: quorum sensing, hierarchical dominance, and reinforcement-based learning. Additionally, it encodes the principle of survival-through-precision, demonstrating how coordinated strategic alignment can outperform structural dominance under resource constraints, inspired by the high hunting efficiency of African wild dogs. PWO integrates three cognitive weighting components: subjective collective preferences (sneeze-based voting), objective data variability (entropy weighting), and experiential reinforcement (pack memory). These are fused via the Mathematical Compromiser, a convex operator that assigns internal trust based on signal stability rather than fixed weighting rules. Applied to European EV gigafactory location selection, PWO reconciled tensions between cost-driven executive preferences and sustainability-based performance indicators, identifying Spain as the most robust alternative. Sensitivity analysis across the dominance spectrum (D=0→1) and multiple episodes confirmed ranking stability without rank reversal. The Markovian update formalizes longitudinal learning for future multi-episode applications. Beyond discrete selection, PWO functions as a diagnostic and competitive resilience mechanism, revealing whether decisions are shaped by leadership authority, structural necessity, historical trends, or precision-based survival logic. It provides a transparent and strategically adaptive architecture for sustainable governance and high-stakes competitive decision environments.

Computers doi: 10.3390/computers15070451

Authors: Baraa Tareq Hammad Ismail Taha Ahmed Mohamed A. Hafez Betty Wan Niu Voon

The proliferation of mobile communications has led to a significant increase in SMS spam, posing challenges related to security, privacy, and user experience. Although numerous machine-learning-based spam detection approaches have been proposed, developing systems that are simultaneously lightweight and resilient to adversarial manipulation remains an open problem. This paper proposes an SMS spam detection framework that incorporates multiple feature extraction methods, including bag-of-words (BoW), Term Frequency–Inverse Document Frequency (TF-IDF), and N-gram models with dimensionality reduction using principal component analysis (PCA), followed by classification using decision tree (DT) and Logistic Regression (LogReg) models. Experimental evaluations on the UCI SMS Spam Collection dataset demonstrate that the TF-IDF-PCA-DT pipeline achieves a detection accuracy of 99% while reducing model size by 77% and inference time by 75%. Robustness evaluation under adversarial text perturbations indicates minimal performance degradation, maintaining an accuracy of 96.5%. These findings demonstrate the practicality of the proposed framework for real-world deployment in resource-constrained environments.

Computers doi: 10.3390/computers15070450

Authors: Edgar Fabián Rivera-Guzmán Vladimir Espartaco Robles-Bykbaev Bernardo J. Vega-Crespo Veronique Verhoeven

Cervical cancer remains a major global public health challenge, and the accurate classification of cervical transformation zones (TZs) constitutes a critical step in early detection and clinical decision-making. However, distinguishing between Type 2 and Type 3 transformation zones remains particularly challenging due to their high morphological similarity and the inherent interobserver variability associated with colposcopic assessment. In this study, we propose a novel Dual-Track Specialist Feature Fusion and Meta-Learning Stacking Ensemble architecture for the automated classification of cervical transformation zones using the Intel & MobileODT Cervical Cancer Screening dataset. The proposed framework integrates a global feature extractor based on ResNet50 (Gatekeeper) with a visual specialist based on InceptionResNetV2, trained exclusively on the most diagnostically ambiguous cases (Type 2 and Type 3). The extracted features are fused and processed through a multi-level stacking scheme composed of Multilayer Perceptron (MLP), Support Vector Machine (SVM), Gradient Boosting (GB), XGBoost, and LightGBM classifiers at the base level, followed by an XGBoost meta-learner and a clinically guided probability calibration strategy designed to maximize diagnostic sensitivity. Experimental results demonstrate a peak overall accuracy of 91.22%, substantially outperforming the baseline ResNet50 model (70%). Furthermore, the proposed system achieved Recall values of 0.90, 0.90, and 0.94 for Type 1, Type 2, and Type 3 transformation zones, respectively, highlighting its ability to accurately identify diagnostically challenging cases. Ablation studies, Grad-CAM visualizations, and external-image validation experiments confirm that the proposed architecture improves discrimination between ambiguous categories, learns clinically meaningful representations, and maintains strong generalization capability across heterogeneous scenarios. These findings demonstrate the potential of visual specialization and calibrated meta-learning strategies for the development of artificial intelligence-assisted colposcopic decision-support systems.

Computers doi: 10.3390/computers15070449

Authors: Andrew N. Shafik Mohamed H. Khafagy Alber S. Aziz Shereen A. Hussein

Accurate instance segmentation of recyclable waste is important for automated sorting and circular economy applications. Mixed metal and glass municipal waste is challenging because metal surfaces are often reflective, while glass objects have transparent boundaries and high visual variability. This paper presents a controlled computer vision framework that uses instance segmentation as an object-level perception layer for mixed metal and glass waste and links eligible detections to screening-level WARM-based environmental interpretation. A seven-class annotated dataset of 1667 images was evaluated using a fixed train/validation/test split. Model development used the validation split for architecture screening, optimization tuning, transfer learning analysis, seed selection, and NMS selection, and the held-out test set was used only for final evaluation. Using mask mAP@50:95 as the primary segmentation metric, the selected YOLOv8m-seg model achieved 0.9435 on the locked test set and obtained a higher mask mAP@50:95 than Mask R-CNN and RF-DETR-Seg under the same locked test protocol. Image-level bootstrap 95% confidence intervals were also used to characterize uncertainty around the locked test segmentation comparison. For environmental reporting, mass-bearing detections were mapped to class-specific mass priors and EPA WARM v16 factors rather than direct mask area-to-mass conversion; the masks were used for instance-level separation, localization, and visual verification, not as direct physical mass measurements. Under the stated assumptions, the framework produced screening-level environmental estimates of 1113.57 MJ of energy savings and 70.20 kg CO2e of avoided emissions, decreasing to 445.67 MJ and 28.62 kg CO2e after applying recycling rate priors. Overall, the framework provides a leakage-aware workflow linking instance segmentation to material-specific WARM-based environmental screening.

Computers doi: 10.3390/computers15070448

Authors: Thanawat Srikaewsiew Sarunya Kanjanawattana Nuntawut Kaoungku Parin Sornlertlamvanich Komsan Srivisut

This paper presents a preliminary technical feasibility study of a landmark-based motion analysis system designed for potential future application in home-based rehabilitation monitoring for frozen shoulder (adhesive capsulitis), developed using computer vision (CV) and human–computer interaction (HCI) principles. The proposed system utilizes real-time body landmark detection to quantify shoulder joint kinematics and provide rule-based automated feedback on exercise execution. The system combines automated and manual components: while shoulder angle assessment, cosine similarity analysis, and keyframe matching are automated, manual researcher input is required to define keyframes corresponding to movement states (start, midpoint, peak) for each therapeutic pose. The CV-driven perception is translated into HCI output, including quantitative movement scores and rule-based feedback indicators, demonstrating the technical potential for objective evaluation of rehabilitation exercise execution without specialized wearable sensors. Technical validation was conducted with 14 healthy volunteers (not frozen shoulder patients) executing standardized shoulder rehabilitation activities, demonstrating shoulder angle measurement with an overall mean absolute error (MAE) of 7.03° against general goniometry and 6.61° against clinical goniometry (RMSE: 8.50° and 8.79°, respectively). Movement similarity classification achieved F1-scores ranging from 0.870 (flexion) to 1.0 (internal rotation) when compared against expert evaluation, though these results are based on a controlled and largely imbalanced dataset with limited incorrect movement examples. The system additionally incorporates a facial expression recognition (FER) module, previously developed and validated in the authors’ prior work, as a supplementary component to support future integration of pain monitoring; this module was not independently validated in the present study. This preliminary technical feasibility study contributes to rehabilitation engineering by demonstrating the potential of semi-automated CV-based motion analysis and rule-based HCI feedback for shoulder movement assessment. The findings indicate technical feasibility for future investigation in home-based exercise monitoring; however, clinical utility cannot be claimed at this stage, as validation with actual frozen shoulder patient cohorts is required.

Computers doi: 10.3390/computers15070447

Authors: Nicolai Iuga Nicoleta Cristina Gaitan Ionel Zagan Vasile Gheorghita Gaitan

As RISC-V processors are increasingly considered for embedded real-time and control-oriented systems, evaluating how timing behavior changes under increasing task concurrency becomes essential. Adding runnable tasks can amplify preemptions, context-switch activity, response-time variability, execution jitter, and deadline pressure. Existing RISC-V simulation and virtual-platform environments mainly target architectural exploration, functional validation, or full-system execution, and do not directly provide a controlled workflow for isolating scheduler-induced timing degradation across large configuration spaces. This paper presents nSim-RV, a configurable and reproducible RISC-V simulation and orchestration framework for scheduler-aware timing scalability evaluation. The framework combines automated campaign generation, deterministic workload configuration, structured dataset aggregation, duplicate validation, and timing-oriented metric extraction. The evaluation compares a standard shared-pipeline execution model with an nMPRA-inspired preserved-context mode under identical scheduler and workload conditions. The campaign includes CoreMark, Dhrystone, and a deterministic synthetic RT-Control workload, 2–32 concurrent tasks, 50 k–1 M cycle observation windows, cache-disabled and cache-enabled configurations, and four-stage and five-stage pipeline organizations, resulting in 864 validated configurations. Results show that increasing task concurrency amplifies timing variability and deadline pressure. Preserved-context execution reduces switching-induced disturbance and delays or reduces higher-pressure timing behavior in several trajectories. Under the five-stage cache-disabled RT-Control configuration at N = 32, it reduces the deadline miss ratio from 3.74% to 2.21%, corresponding to a 41.1% relative reduction, with the clearest benefits observed for Dhrystone and RT-Control at intermediate–high task counts.

Computers doi: 10.3390/computers15070446

Authors: Daniel Asimionesei Nirvana Alina Popescu

Cellular networks from 2G to 5G have evolved to improve performance, reliability, and security. However, the protection of user identity and location remains a persistent challenge across generations. Although newer architectures introduce stronger authentication, temporary identifiers, and privacy-preserving mechanisms, attacks such as IMSI catching, paging-based tracking, downgrade attacks, and identifier correlation continue to affect cellular systems. This study analyzes identification and location attacks in 2G, 3G, 4G, 5G Non-Standalone (5G NSA), and 5G Standalone (5G SA) networks. The analysis focuses on studies published between 2018 and 2026 and follows a structured methodology based on research questions, inclusion and exclusion criteria, quality assessment, and comparative synthesis of studies identified in scientific databases. Following the selection process, 86 studies were included in the final analysis. The review compares attack vectors, affected identifiers, exploited procedures, adversary capabilities, reported performance metrics, and proposed mitigations for each network generation. The results show that vulnerabilities inherited from previous generations, especially GSM/2G, remain relevant in modern architectures through downgrade attacks, fallback mechanisms, and LTE anchoring in 5G NSA deployments. The main contribution of this study is a cross-generational comparative synthesis of identity and location attacks in cellular networks. The findings highlight the need for rigorous standards implementation, effective detection mechanisms, reduced reliance on legacy technologies, and practical evaluation of mitigation solutions in real cellular environments.

Computers doi: 10.3390/computers15070445

Authors: Mohammed Aqeel Ismail Okuthe P. Kogeda

The increasing complexity and volume of mobile network traffic present significant challenges to maintain consistent Quality of Service (QoS) across diverse applications. Accurate traffic classification enables application-aware resource allocation by distinguishing applications with different bandwidth, latency, and reliability requirements. Traditional classification techniques, including port-based identification and Deep Packet Inspection (DPI) have become inadequate and less effective due to widespread encryption, port masquerading, and growing privacy concerns. This paper presents a supervised learning-based approach for application-level network traffic classification specifically as a foundation for QoS optimization in future 5G networks. Since publicly available labeled 5G traffic datasets remain limited, this study uses the MIRAGE-2019 mobile traffic dataset as a proxy dataset to evaluate the proposed classification framework. A Random Forest classifier was implemented using flow-level statistical features extracted from the mobile application traffic. The framework further incorporates a rule-based QoS policy mapping informed by RFC 4594 DiffServ service class guidelines to assign application-specific priority levels, bandwidth requirements, latency sensitivity, and jitter tolerance. Experimental evaluation achieved an overall classification Accuracy of 71.83%, a Macro F1-score of 0.6701, and a Weighted F1-score of 0.7227 across twenty mobile applications. Although the experiments were conducted using a pre-5G mobile traffic dataset, the results demonstrate that supervised machine learning can effectively classify encrypted mobile application traffic and provide a practical foundation for application-aware QoS policy enforcement in future 5G and next-generation mobile networks.

Computers doi: 10.3390/computers15070444

Authors: Madina Ipalakova Dana Tsoy Sanzhar Otkilbayev Yevgeniya Daineko Danil Sharipov Umitkhan Turzhanov

The timely integration of technology in education is a key factor in developing competitive specialists. Given this, the development of a digital educational platform for training engineering specialists is of strategic importance. The paper presents the development of a proprietary digital educational platform using immersive technologies for training engineering specialists while providing educators with easier access to their educational progress and ability to spend less time processing it. The results are considered data from an initial test, aimed primarily at assessing user perception, usability, and the potential of the immersive environment. The platform’s effectiveness stems from its ability to simulate complex processes and allow students to independently control the experiment.

Computers doi: 10.3390/computers15070443

Authors: Marianna Thode Ioannis Paliokas

Education about AI, including its opportunities and limitations, is essential for responsible academic development. It helps students, teachers, parents, researchers and practitioners understand both the strengths and the limitations of generative AI and ensures that it is applied in ways that are ethical and socially responsible. Without this knowledge, there is a risk that generative AI will be misused or misunderstood, reducing its potential benefits. This study evaluates an educational game designed to integrate generative AI in elementary education (in art lessons), aiming to create awareness regarding the limitations of generative AI. Data was collected from 204 game sessions in two primary education schools in Greece, using serious games for art education, while the accuracy of AI outputs and the usability and level of user satisfaction were recorded. Following these tasks, researchers used the concurrent think-aloud (CTA) protocol to identify specific usability issues and responses of elementary students in biased, incomplete or inaccurate AI results. Experimental findings from this small-scale exploratory pilot study (16 students) indicated that combining RTA with serious games in art courses successfully reveal how children understand AI outputs and evaluate them based on their initial expectations.

Computers doi: 10.3390/computers15070442

Authors: Thananchai Khamket Jatuphum Juanchaiyaphum Theeraya Uttha Manasawee Kaenampornpan Jantima Polpinij

Bug severity classification supports software maintenance by helping developers prioritize defect resolution. Most existing approaches combine bug report titles and descriptions into a single representation, often assuming that both textual components contribute equally to severity prediction. However, concise titles and detailed descriptions may provide different types of severity-related information. This study investigates their relative contribution and proposes an Adaptive Title–Description Fusion (ATDF) framework that separately encodes title and description representations and adaptively estimates their contribution during feature fusion. Experiments were conducted on Mozilla Bugzilla repositories using a multiclass classification setting with five severity categories. The proposed framework was compared with title-only, description-only, direct concatenation, and standard gated fusion baselines. ATDF achieved the highest overall performance, with a Macro-F1 score of 0.771, compared with 0.756 for standard gated fusion and 0.739 for direct concatenation. Additional analyses showed that title and description contributions varied across severity categories and repositories under the current experimental setting. Overall, the findings suggest that considering the relative contribution of title and description information can provide a practical approach to improving bug severity classification while offering additional insight into title–description fusion behavior in software repositories.

Computers doi: 10.3390/computers15070441

Authors: Fan Yang Lixin Lyu

To address the difficulty of accurately capturing the dynamic patterns of carbon emission time series—characterized by nonlinearity, non-stationarity, and complex fluctuations—this paper proposes a carbon emission prediction model based on an elite-guided collaborative social spider learning optimization algorithm (EGC-SSLO) integrated with a Long short-term memory (LSTM) network. First, considering the limitations of the standard stochastic social learning optimization (SSLO) algorithm in complex high-dimensional optimization problems, such as insufficient elite information guidance, weak local exploitation in the later stages, and a tendency to become trapped in local optima, three complementary improvement strategies are introduced. The adaptive elite mean-guided search strategy enhances the search directionality by incorporating the cooperative information of the best individual and the elite mean. The worst-individual hybrid Cauchy–Lévy search mechanism achieves a dynamic balance between early-stage global exploration and late-stage local exploitation through long-range Lévy flights and fine-grained Cauchy perturbations. The quadratic directional exploitation strategy further refines the search trajectory of candidate solutions, thereby improving convergence accuracy. These three strategies significantly enhance the optimization performance without increasing the time complexity order of the algorithm. Experimental results on the CEC2017 (30-dimensional), CEC2020 (20-dimensional), and CEC2022 (20-dimensional) benchmark suites demonstrate that EGC-SSLO consistently outperforms classical algorithms such as PSO, GWO, and HHO, as well as their improved variants, in terms of convergence accuracy, convergence speed, and robustness. Furthermore, the Wilcoxon rank-sum test and Friedman test confirm that the observed improvements are statistically significant. Finally, an EGC-SSLO-LSTM carbon emission prediction model is constructed and applied to daily carbon emission data in China from 2019 to 2025 for empirical analysis. The experimental findings show that the EGC-SSLO-LSTM model markedly outperforms both the standard LSTM and SSLO-LSTM approaches across key evaluation metrics, including mean absolute error (MAE), root mean square error (RMSE), and the coefficient of determination (R2). In particular, the MAE is decreased by 39.9% and 4.64% compared with the two benchmark models, respectively, which highlights the strong effectiveness and practical potential of the proposed method in real-world carbon emission forecasting applications.

Computers doi: 10.3390/computers15070440

Authors: Mina Soltani Siapoush Jim Alves-Foss

RISC-V’s open and extensible design improves flexibility, but it also permits vendors to customize trap behavior and reserved opcode space in ways that can affect exception handling. This paper presents a scoping review of RISC-V exception-handling security, focusing on how traps, privilege transitions, control registers, and handler routines can become attack surfaces when they are not correctly protected. We organize the literature by a five-stage exception lifecycle and map documented attacks and defenses to the architectural state they affect. Our review shows that exception-handling vulnerabilities arise from exposed control registers, insufficiently protected trap entry and return paths, and flawed handler execution. Recent attacks, such as GhostWrite, halt-and-catch-fire conditions, and side-channel exploits, illustrate how weaknesses in exception management can lead to privilege escalation, denial of service, or data leakage. In contrast, defenses are unevenly distributed across the lifecycle: backward-edge return-time integrity is addressed by multiple dedicated mechanisms, while dispatch-time and privilege-transition protection remain largely indirect and are often embedded within broader trusted-execution designs. Because the primary studies report heterogeneous cost metrics, the overhead values we extract are presented as reported rather than normalized across a common benchmark. The literature suggests that lightweight mechanisms, such as Physical Memory Protection and software-based control-flow enforcement, generally incur modest overhead, whereas stronger approaches, such as CHERI-RISC-V and Trusted-Execution Environments, may impose substantially higher costs on exception-heavy workloads. Overall, our findings indicate that exception handling should be treated as a security-critical boundary and that future work should emphasize stage-specific defenses, formal verification, and tighter integration between hardware and software protections.

Computers doi: 10.3390/computers15070439

Authors: Luís Nogueira Jorge Coelho

Fault tolerance is a fundamental requirement of distributed systems, and actor-based runtimes provide a widely adopted approach for building resilient and highly concurrent applications. Although several actor ecosystems offer mechanisms for supervision, failure detection, and recovery, comparative studies frequently focus on performance metrics rather than fault-tolerance behaviour. This paper presents a language-independent benchmarking framework for evaluating fault tolerance in actor-based runtimes. The framework was implemented using three representative ecosystems: Elixir/BEAM, Scala/Akka, and Go/Proto.Actor. A distributed chat-based benchmark application was used to measure throughput, reconnection latency, and failure-detection latency under recurring transient failures. All implementations followed an equivalent architecture and were executed under identical experimental conditions. The study deliberately targets a single, well-defined fault model: the supervised crash recovery of in-memory, effectively stateless actor services, in which chat actors are abruptly terminated and restarted by their supervisors while clients rediscover and reconnect to them. Stateful recovery (actor state, mailbox contents, in-flight or persistent messages), as well as multi-node network effects, are explicitly out of scope. Accordingly, the benchmark characterises supervised crash–recovery behaviour for largely stateless actor services rather than providing a comprehensive evaluation of actor-based fault tolerance. The results reveal distinct trade-offs among the evaluated ecosystems. Elixir achieved the highest throughput and the lowest throughput variability under fault conditions, while Scala/Akka consistently provided the lowest reconnection and failure-detection latencies, particularly at large scale. Go/Proto.Actor remained competitive in throughput-oriented scenarios but showed greater degradation in recovery-related metrics as concurrency increased. The results indicate that no single runtime dominates all evaluated dimensions of recovery behaviour. Beyond the runtime comparison, this work contributes a reproducible benchmarking framework that provides a foundation for future empirical studies of actor-based runtime recovery under controlled fault conditions.

Computers doi: 10.3390/computers15070438

Authors: Cheslin van Wyk Taryn Michael Colin Chibaya

Water scheduling is a complex optimization problem that requires efficient and adaptive solution approaches. Metaheuristic techniques, particularly swarm intelligence models, have increasingly been applied to address these challenges. This study presents a scoping review that maps and synthesizes the existing literature on the application of swarm intelligence in water scheduling. Guided by the PRISMA-ScR framework and the JBI Population–Concept–Context (PCC) model, relevant studies published between 2015 and 2025 were identified across multiple databases. From an initial pool of 1357 studies, only 23 met the inclusion criteria and were subjected to detailed analysis. The findings reveal a strong concentration of research on water distribution networks, coupled with limited methodological diversity across the reviewed studies. There is an absence of explicit focus on resource-constrained or arid environments contexts where water-scheduling challenges are often most acute. Geographically, the literature is heavily skewed toward Asia, with the majority of studies conducted in China (n = 7) and Iran (n = 6). In contrast, only one study originated from Africa and one from Australia despite the disproportionate severity of water scarcity challenges across the African continent. The review exposes a critical gap in the literature and underscores the need for more context-aware, hybrid swarm intelligence models that explicitly account for the socio-economic and environmental constraints of water-stressed regions.

Computers doi: 10.3390/computers15070437

Authors: Weichen Huang Sihao Dong Zhiheng Fan Xiaohu Zhang

Synthetic aperture radar (SAR) ship detection is critical for maritime surveillance; however, accurately identifying targets in complex marine environments remains a persistent challenge due to severe sea clutter and coastal interference. To address the prevalent issues of missed detections and false alarms, this paper proposes a novel deep learning framework named AFN-YOLO (Adaptive Frequency Network–You Only Look Once). Specifically, we propose a C2f_FF (C2f Frequency Fusion) module that dynamically extracts and fuses frequency-domain and spatial-domain information, effectively suppressing background interference and artifacts from nearby buildings. Additionally, a TAS-FPN (Triplet Attention-based Spatial FPN) architecture is integrated to capture multiscale features, significantly improving the detection capability for small and overlapping ship targets. Furthermore, the loss function is optimized to compel the model to focus on salient target features while disregarding irrelevant background data. Extensive experiments on the SAR Ship Detection Dataset (SSDD) and the High-Resolution SAR Images Dataset (HRSID) validate the effectiveness of our approach.

Computers doi: 10.3390/computers15070436

Authors: Syed Sadiqur Rahman

Lightweight, privacy-aware and adversarial robust intrusion detection is required for the proliferation of Internet of Things (IoT) devices. In the Industrial Internet of Things (IIoT), centralized detectors can be compromised by adversarial perturbations via gradient-based attacks, making them susceptible to raw traffic. We suggest Federated Learning-Adaptive Gated Recurrent Unit (FL-AdGRU), a Federated approach that combines a lightweight Gated Recurrent Unit (GRU) classifier with alternating adversarial fine-tuning on each client using FGSM and PGD, without any communication overhead. A two-stage resampling scheme (UCAS-SMOTE) reduces the class-imbalance ratio from 4081:1 to ≈4:1, followed by 61 features being reduced to 40 by a mutual-information selector (MI-SelectK). Under this scenario, FL-AdGRU achieves 99.9% accuracy and 0.999 weighted F1 (+6.5 p.p. over the federated DNN baseline), with no loss of accuracy when facing clean attacks, and boosts Fast Gradient Sign Method FGSM/Projected Gradient Descent (PGD) robustness by +19.3/+19.0 p.p. at the same level of ϵ = 0.1, thus effectively balancing the accuracy–robustness trade-off. It is robust (97.8%/84.2% on UNSW-NB15) and generalizes well to UNSW-NB15, while decaying slowly in skeptical scenarios (≈99.9% weighted F1 for moderate skew, 93.9%/86.7% for severe). Assuring data-locality privacy through exchange of only model weights; defenses against inference attack are left for future work. FL-AdGRU, with a total communication of 43.8 MB (≈50× less than centralized training), is deployable on bandwidth-constrained IIoT networks.

Computers doi: 10.3390/computers15070435

Authors: Kexuan Liu Yanyu Chen Ying Wang Yuxiang Zhou Tao Wan Xin Xie

Resource-constrained dynamic heterogeneous redundancy (DHR) systems use executor diversity and runtime reconfiguration to reduce stable attack surfaces. However, effective scheduling cannot rely only on heterogeneity or movement frequency, because repeated exposure, shared vulnerability sources, service disturbance, and switching overhead jointly shape executor-subset selection. This paper proposes RACS, a risk-aware cost-constrained scheduling method for resource-constrained DHR systems. RACS evaluates candidate subsets by jointly considering heterogeneity, historical confidence, readiness, common-vulnerability risk, exposure memory, and switching cost. We evaluate RACS using a controlled simulation protocol covering multiple scheduling principles and attacker behaviors, including common-vulnerability pressure, burst-adaptive exploitation, and adaptive target selection based on observed scheduling patterns. The results show that RACS does not optimize a single metric in isolation, but maintains a consistent security–cost trade-off. It reduces common-vulnerability risk and switching cost in common-vulnerability settings, reduces burst-triggering high-risk states under adaptive pressure, and maintains competitive risk–cost behavior when attackers adapt to historical scheduling behavior. Robustness and scalability analyses clarify the effects of vulnerability-family estimation errors, parameter choices, and executor-pool size. These findings provide controlled simulation evidence for joint risk–cost modeling in DHR executor-subset scheduling, while testbed validation remains future work.

Computers doi: 10.3390/computers15070434

Authors: Robert Campbell

Agentic AI systems depend on classical public-key cryptography for agent identity, tool invocation, inter-agent communication, model integrity, and persistent state, exposing them to a cryptographically relevant quantum computer (CRQC) along two axes: confidentiality (harvest-now-decrypt-later) and integrity (harvest-now-forge-later). Existing post-quantum migration guidance addresses static, operator-controlled enterprise estates, while emerging agent-identity work omits post-quantum cryptography entirely; neither treats non-human-identity-dense, runtime-negotiated agentic systems as a distinct migration class. This paper develops a conceptual framework that does. It organizes agentic cryptography into seven migration surfaces and separates each identity into a credential layer (symmetric, operator-held, low-risk) and a trust-anchor layer (the asymmetric roots that underwrite the fleet). These layers scale inversely: a small set of trust anchors carries a forge-later blast radius equal to the population beneath it, so migration effort and forge-later risk rank the work in opposite orders. A migration matrix and a parametric effort-and-risk model formalize this, yielding the core sequencing rule: migrate anchors first. Because agentic adoption is ongoing, it also reframes migration from a finite inventory into a continuously regenerating problem, distinguishing remediation of the installed base from prevention of new classical-cryptographic debt in future deployments. It closes with oversight and procurement implications for federal post-quantum readiness.

Computers doi: 10.3390/computers15070433

Authors: Chrysovalantis Kefalis Constantine (Kostas) Skordoulis Chara Papoutsi Athanasios Drigas

Universal Design for Learning (UDL) is increasingly recognized as a critical framework for promoting equity and accessibility in Science, Technology, Engineering, and Mathematics (STEM) education. In this review, we synthesize research published between 2002 and 2025 on UDL in STEM education, with attention to technology-enhanced learning environments. Searches were conducted in Scopus, Web of Science, Education Resources Information Center (ERIC), and IEEE Xplore. After title/abstract screening and full-text assessment, 13 studies were included: 11 empirical studies and 2 theoretical/conceptual papers. Across the empirical studies, the findings suggest promising but heterogeneous evidence that UDL-informed STEM practices may support access, engagement, self-regulation, persistence, and inclusive participation, while evidence for standardized achievement gains remains more mixed. The synthesis also highlights recurring design considerations, including multimodal representation, flexible participation, scaffolded learning, and social–emotional support. We conclude that UDL offers practical guidance for inclusive STEM education, particularly when learning environments are designed to support cognitive access, emotional engagement, collaboration, and diverse learner needs.

Computers doi: 10.3390/computers15070432

Authors: M. Sirajul Islam

This paper investigates how institutional fragilities shape cross-border cyber risk in financial systems, using the 2016 Bangladesh Bank heist as an illustrative case from a developing-country context. Drawing on a scoping review of the multidisciplinary literature on cybersecurity governance, financial cybercrime, and cross-border cyber risk, combined with an exploratory case study, the Bangladesh Bank heist is explored through a multi-level analytical lens encompassing macro-level institutional and geopolitical factors, meso-level ecosystem interdependencies, and micro-level organizational vulnerabilities. The analysis suggests that the 2016 heist was not solely a technical intrusion but was also associated with broader governance, regulatory, infrastructural, and organizational weaknesses. Fragmented governance arrangements, digital dependencies, limited operational resilience, inadequate cybersecurity controls, and human-factor vulnerabilities emerged as important dimensions of risk. Comparison with selected incidents from other developing and emerging economies indicates that similar patterns may occur across different institutional settings, although their manifestation remains context dependent. Rather than advancing a general theory, the highlights the importance of strengthening governance capacity, regulatory coordination, and cyber resilience within globally interconnected financial systems.

Computers doi: 10.3390/computers15070431

Authors: Ashwani Kumar Dubey Divya Upadhyay Alvaro Rocha Ayush Goyal

The combination of predictive analytics and blockchain-technology-enabled smart contracts has the revolutionary potential to transform logistics network operations. This paper presents a five-layer architecture to improve logistics network management in terms of efficiency, minimize costs, and enable data-driven decision making capabilities. The proposed system includes collection of data, blockchain-enabled smart contracts, integration of the system, and real-time monitoring. With rigorous performance analysis, this paper presents phenomenal gains in predictive accuracy, transactional security, and system operational efficiency. It also proposes real-time monitoring features that ensure continuous adjustment to changing conditions, thereby providing a reliable solution to contemporary logistics network challenges. This work also bridges research gaps, showing practical applications and empirical evidence in favor of the integrated solution to establish its effectiveness. Experimental evaluation demonstrates strong predictive performance (R2 up to 0.96), successful blockchain-based transaction execution, and transportation cost reduction through optimized allocation. The transportation linear programming model reduced shipping costs from INR 1080 to INR 854 through optimal allocation across sources and destinations.

Computers doi: 10.3390/computers15070430

Authors: Yahya Masri Emily Ma Zifu Wang Joseph Rogers Chaowei Yang

System logs are essential for monitoring and diagnosing modern computing infrastructure; however, their scale and complexity require reliable automated interpretation. Although severity levels are predefined metadata, treating classification as an end task provides limited insight into a model’s ability to understand logs. Instead, severity classification can serve as a benchmark for probing runtime log comprehension. Using real-world journalctl data from Linux production servers, this study evaluates nine small language models (SLMs) and small reasoning language models (SRLMs) under zero-shot, few-shot, and retrieval-augmented generation (RAG) prompting. The results reveal clear performance stratification. Qwen3-4B achieves the highest accuracy (95.6%) with RAG, while Gemma3-1B improves substantially from 20.25% to 85.28%, and Qwen3-0.6B reaches 88.12% despite weak baseline performance. In contrast, several SRLMs exhibit performance degradation when paired with RAG. Efficiency further differentiates the models. Most Gemma and Llama variants complete inference in under 1.2 s per log, whereas Phi-4-Mini-Reasoning requires more than 228 s while achieving less than 10% accuracy. These findings indicate that architectural design, training objectives, and the ability to integrate retrieved context jointly determine performance. Overall, the results support severity classification as a practical lens for evaluating model competence and real-time deployability in digital twin systems.

Computers doi: 10.3390/computers15070429

Authors: Obai Alashram Obada Al-Khatib Abeer Elkhouly

Deep multiscale learning has emerged as a promising venue for robust image detection and multi-object tracking in adverse conditions, but the current solutions tend to be impacted by the issues of occlusion, scale variation, and background clutter, focusing on each of them separately and restricting the generalization. In a direction to address these gaps, this piece of writing proposes a unified model that incorporates HRNet to extract high-resolution features, DETR to make use of transformers for detection, and TrackFormer to identify in an identity-preserving manner. Data was based on the MOT17 benchmark dataset, which provides various urban video sequences, including annotated bounding boxes and identities, to guarantee a test that is rigorous. The approaches were selected due to their complementary advantages: HRNet keeps fine-grained spatial information, DETR allows us to locate the objects in an accurate way, and TrackFormer tracks the trajectories across fragments. Experiments show good performance, with a mean detection AP of 70.9, precision of 76.5, recall of 72.8, MOTA of 74.8, IDF1 of 70.2, and HOTA of 63.6, maintaining real-time performance of 26 FPS with a latency of 38.5 ms per frame. In general, this work offers a globally scalable, end-to-end system for problems like surveillance and self-driving, and future work aims to address outrageously dense scenes, enhance cross-dataset generalization, and come up with lightweight systems to deploy these edges.

Computers doi: 10.3390/computers15070428

Authors: Diego Armando Pérez-Rosero Andres Camilo Lopez-Boscan Andrés Marino Álvarez-Meza David Augusto Cárdenas-Peña German Castellanos-Dominguez

Interpretability is essential for deploying deep learning (DL) models in electroencephalography (EEG)-based neurotechnology, particularly in brain–computer interfaces and clinical decision-support settings. Existing post hoc explainable artificial intelligence (XAI) methods often yield single-domain attribution maps, limiting their capacity to characterize the joint spatial, temporal, and spectral structure of EEG dynamics. In addition, perturbation-based strategies may disrupt physiological signal organization, whereas gradient-based methods require access to model internals and are therefore tied to specific classifier architectures. Here, we introduce space–time–frequency KernelSHAP (STF-KernelSHAP), a model-agnostic Shapley framework for physiologically coherent EEG explainability. The method comprises three stages. First, EEG trials are decomposed into structured channel–time–frequency cells using segment-wise spectral analysis. Second, coalitions are formed over complete channel–time–frequency cells and reconstructed in the signal domain to support physiologically informed perturbations. Third, class-conditional relevance is estimated with a KernelSHAP-based weighted surrogate model that uses only model outputs, enabling architecture-independent Shapley estimation. We evaluate STF-KernelSHAP on two prerecorded public datasets: the GIGA motor imagery/movement execution (MI-ME) dataset for motor imagery (MI) decoding and the IEEE DataPort EEG Data for Attention-Deficit/Hyperactivity Disorder (ADHD)/Control Children dataset for ADHD detection. For ADHD detection, the T-GARNet base classifier interpreted with STF-KernelSHAP achieved 73.33% accuracy and 79.86% area under the curve (AUC); these values characterize classifier performance rather than the explainer itself. We compare the framework against KernelSHAP, local interpretable model-agnostic explanations (LIME), Occlusion, Integrated Gradients, and gradient-weighted class activation mapping++ (Grad-CAM++). Fidelity is assessed with Deletion and remove and debias (ROAD), while qualitative analyses examine topographic and frequency-band attribution maps. Results show that STF-KernelSHAP remains functionally competitive with established XAI methods while providing window-dependent and frequency-specific explanations. Overall, STF-KernelSHAP offers a physiologically informed and model-agnostic alternative for multidomain EEG interpretability.

Computers doi: 10.3390/computers15070427

Authors: Pongsathon Pookduang Wirapong Chansanam

Purpose/Background: Small and medium-sized enterprises (SMEs) that deploy Enterprise Resource Planning (ERP) systems face a persistent paradox: although ERP centralises organisational data, frontline users frequently lack the technical expertise to navigate complex menu structures, preventing efficient information retrieval for decision-making. This study presents a preliminary Research and Development (R&D) effort to design, build, and qualitatively evaluate a prototype intelligent question-answering system that connects to Odoo ERP through application programming interfaces (APIs) via Retrieval-Augmented Generation (RAG), enabling natural-language access to real SME business data. Methods: A single-organisation R&D case study was conducted at an SME in Khon Kaen Province, Thailand. The development cycle comprised problem analysis and requirement specification, system architecture design, prototype construction, integration and deployment, iterative testing and refinement, and multidimensional evaluation. The prototype was implemented with Chainlit (conversational interface), FastAPI (orchestration and tool-calling layer), and Odoo XML-RPC/JSON-RPC APIs (structured data retrieval). A fixed set of 20 test questions spanning four complexity levels (easy, moderate, complex, out-of-scope) was evaluated by three automated tools (OpenAI Evals, DeepEval, Ragas), by real-task verification against live ERP data, by five domain experts using 5-point Likert-scale questionnaires, and by three end users from the case-study organisation, who additionally completed the System Usability Scale (SUS). Given the very small expert and user samples, all human evaluation results are reported descriptively as preliminary, exploratory indicators rather than as statistically generalisable measures. Results: Automated evaluation achieved indicative pass rates of 95.00% (OpenAI Evals, 19/20), 90.00% (Ragas, 18/20), and 85.00% (DeepEval, 17/20). Descriptive expert feedback (n = 5) yielded an overall mean of 3.82 (high level), and descriptive end-user feedback (n = 3) yielded an overall satisfaction mean of 4.33 (highest level). The SUS score was 66.67/100, sitting at the boundary between ‘OK’ and ‘Good’ and revealing a divergence between high stated satisfaction and lower confidence in independent system use (item 9, raw mean = 2.33) and a stronger perceived need for expert assistance (item 4, raw mean = 2.67). These results are interpreted as preliminary diagnostic signals for further development rather than as confirmatory evidence. Conclusions: This preliminary R&D study suggests that a RAG-based ERP chatbot can meaningfully simplify ERP data access for SME users, while exposing persistent gaps in multi-step reasoning, user confidence, and data privacy boundaries that must be addressed in subsequent development cycles. The SUS pattern, in particular, suggests a ‘novelty effect’ in which users are enthusiastic about the natural-language interface yet remain anxious about correctness and stability during real tasks. Originality/Value: This work contributes a transparent, replicable preliminary R&D blueprint that combines (i) live API-mediated RAG over structured ERP data, (ii) a complementary multi-tool automated evaluation set (OpenAI Evals + DeepEval + Ragas), (iii) descriptive expert and end-user feedback, and (iv) SUS-based usability assessment, all documented for a single Thai SME using Odoo 18. The study explicitly positions itself as an early step toward larger, multi-site, and on-premise deployments of trustworthy ERP-integrated conversational agents.

Computers doi: 10.3390/computers15070426

Authors: Abdel Rahman Alkharabsheh Ghaya Binsalma Mahra Alharmi Ruqia Alshateri Shahad Altaee Mousa Sweidan

Insider threats continue to be the most persistent and most destructive threat to cybersecurity; malicious or negligent users work only in the real-time restricted area of the organization and are gradually breaking the boundaries of company norms. Conventional rule-based and statistical detection methods have difficulty detecting inconspicuous, context-dependent, and ever-changing behavior, leading to detection delays and high false-positive rates. Our paper introduces an explainable AI-based Insider-Threat Detection (AIB-ITD) model that integrates enterprise telemetry—including email, web, logon/VPN, and file events—into a unified behavioral framework. The effectiveness of combining heterogeneous behavioral indicators observed in AIB-ITD is consistent with recent behavioral analytics implementations that have demonstrated the value of multimodal user-behavior profiling for insider-threat identification in enterprise environments. The proposed AIB-ITD framework is based on anomaly-driven processing, unsupervised models (Isolation Forest, PCA reconstruction, and Autoencoder) are combined with sequential modeling (with an LSTM Autoencoder) to model both static and temporal deviations in behavior. An ensemble strategy is applied to combine the outputs of these models to yield a probabilistic insider risk score. To improve transparent analysis and to help the analyst gain trust, SHapley Additive Explanations (SHAP) is used to keep every detection outcome transparent and interpretable using the features. It also integrates feature correlation analysis, static vs sequential-model comparisons, and SHAP stability assessment to validate methodological robustness and reproducibility. An experimental review of the hybrid ensemble using the SEI/CMU CERT Insider Threat Dataset reveals that it performs better than single models for anomaly detection and stability, especially with the inclusion of temporal patterns. The assessment prioritizes anomaly score consistency and reliable risk ranking, rather than classification accuracy, to better reflect real deployment scenarios. In addition, an Automated Response and Privilege Containment (ARPC) feature automatically converts risk scores to multilevel mitigation actions that serve to protect the privacy of the user as the least privileged policies are enforced promptly. The proposed model showed superior robustness, stability, and operational effectiveness to classical methods, especially in the presence of scarce labeled data. Through hybrid anomaly recognition, explainable AI and automated response, AIB-ITD is a practical and scalable solution for next-generation insider-threat detection in enterprise systems.

Computers doi: 10.3390/computers15070425

Authors: Alberto Gudiño-Ochoa Harold Felipe Calderón-González

Forecast reconciliation has been widely studied in cross-sectional and temporal hierarchies, but its role in cross-temporal settings for photovoltaic (PV) forecasting remains insufficiently examined. In particular, the relative benefits of reconciliation across heterogeneous forecasting approaches, including statistical, machine learning, deep learning, and foundation models, have not been clearly established. This study addresses that gap by evaluating direct, univariate, and iterative cross-temporal reconciliation strategies applied to TBATS, LightGBM, KAN, NBEATSx, NHITS, and TimeGPT using Belgian PV generation data from 2020 to 2025 across weekly, daily, and hourly frequencies and national, regional, and provincial levels. Model efficacy is assessed through 52-week walk-forward cross-validation, which provides a full-year coverage. Under the fixed-configuration experimental protocol adopted in this study, the results show that the gains from reconciliation vary substantially across forecasting families. LightGBM achieved the largest observed gains, with its univariate and iterative schemes achieving global error reductions of up to 19.6% relative to the Bottom-Up benchmark. KAN, NHITS, and NBEATSx also benefited from reconciliation, with their best reconciled variants yielding reductions of up to 11.9%. TimeGPT and TBATS achieved reductions of up to 9.2% and 14.5%, respectively, although their global errors were higher than those obtained by the best machine learning and deep learning configurations in this evaluation. Across the fixed baseline configurations considered here, LightGBM obtained the lowest global errors before and after reconciliation. These findings show that cross-temporal reconciliation can be an effective post-processing strategy, but its impact depends strongly on the underlying base forecasting model. Therefore, the observed advantage of LightGBM should be interpreted as conditional on the adopted feature set, implementations, and baseline configurations.

Computers doi: 10.3390/computers15070424

Authors: Hongji Zhu Liping Chen Shuguang Han

The nesting and cutting of thin materials are critical processes in industrial manufacturing, often involving multi-layer stacking to optimize production efficiency. However, material defects complicate the process, requiring optimization of both layout and defect avoidance under multi-layer heterogeneous constraints. Moreover, existing methods struggle with large search spaces and high computational complexity, limiting their industrial applicability and affecting material utilization and machining accuracy. To address these challenges, we propose HGATrans-MNCD, an intelligent nesting optimization algorithm that integrates defect avoidance and material waste reduction. Initially, areas with high defect overlap are prioritized using an enhanced No-Fit-Polygons strategy to ensure global defect avoidance. A heuristic approach is then employed to optimize the initial nesting sequence. Subsequently, a Transformer-based module leverages prior knowledge to efficiently perturb and refine the sequence, facilitating global optimization. Experiments on a benchmark dataset of multi-layer defect scenarios demonstrate that HGATrans-MNCD effectively addresses irregular defect patterns, enhancing material utilization by 2–8%. Our algorithm performs especially well in scenarios involving spatially coupled defects, offering a novel solution to complex multi-constraint optimization problems.

Computers doi: 10.3390/computers15070423

Authors: Artrim Kjamilji

We tackle the problem of secure and private feature selection by homomorphically evaluating features’ information gains over the encrypted data of horizontally partitioned private datasets owned by edge IoT (Internet of Things) devices. In the process, we use a powerful cloud server to do the bulk of the costly homomorphic encryption aggregations. We proceeded with secure and private federated learning (training) and Machine Learning (ML) classification over the selected features in the same environmental settings (context). In the process, the participants interact with each other under strict security, privacy, and efficiency requirements. To this end, to each participant’s interaction we provide confidentiality, integrity, and authenticity (CIA) by signing its hashed contents with the corresponding participant’s private key. We assure consistency among interactions by introducing timestamps and linking them with the hashed content(s) of the preceding interaction(s). Those linked blocks of hashed content(s) from each interaction of participants while running the protocols produce the so-called chain-of-blocks (COB) structure, which will be utilized to detect malicious edge IoT dataset owners, unauthorized participants, and network errors. The security of the proposed protocols is proven through rigorous mathematical modeling. Extensive experimental evaluations over benchmark datasets give an advantage to our secure protocols ranging from several times to orders of magnitudes w.r.t to the state of the art in terms of computation and communication costs, as well as security and privacy characteristics. Moreover, since the utilized underlying cryptographic techniques are resilient to quantum computer attacks, the proposed algorithms are applicable to the post-quantum world.

Computers doi: 10.3390/computers15070422

Authors: Zhijun Shen Qian Guo Kaiyuan Pang Xinlei Cai Zhenfan Yu Kunhao Feng Tao Yu

Voltage and reactive power control in active distribution networks is a safety-critical and highly dynamic problem, where traditional optimization methods often struggle to balance efficiency and robustness under complex operating conditions. Recently, large language models (LLMs) have shown promise in sequential decision-making tasks, but their direct application to power system control remains limited by the lack of physical grounding and safety guarantees. In this paper, we propose SafeVolt, a closed-loop LLM-based framework that integrates multi-candidate action generation, simulator-in-the-loop evaluation, and a fine-tuned expert judge for safety-aware decision making. In addition, a high-level rule distillation mechanism that converts successful control experiences into reusable operational axioms is introduced to enable iterative self-improvement. Experiments on a standard distribution network scenario demonstrate that the proposed method outperforms representative baselines, achieving substantial improvements in average reward, voltage violation rate, reactive power loss, and system stability. In particular, voltage violations and extreme events are substantially reduced, indicating enhanced operational safety. These results suggest that combining LLM reasoning with physical simulation and structured feedback provides a promising direction for reliable and adaptive power system control.

Computers doi: 10.3390/computers15070421

Authors: Safa Elkefi Salma Bhar Achraf Tounsi Duxiao Hao

Immersive solutions are becoming a trending technology for decision support across fields such as transportation, healthcare, and urban planning. Despite their role, the mechanism by which they affect decision-making is unclear. Our study examines the design primitives in immersive technology that are manipulated to influence decision-making and synthesizes how they operate to shape decision outcomes. We follow PRISMA guidelines to search. A total of 198 studies were included. Eight primitive families were identified, including perceptual realism, environmental structure, interactivity, temporal simulation, embodiment, social presence, multisensory integration, and other contextual manipulations. Mechanisms through which they impacted decision-making were classified into cognitive, perceptual, affective, motivational, social-influence, and behavioral-heuristic mechanisms. Perceptual realism, environmental structure, and interactivity emerged as the most frequently investigated primitives, while presence, risk perception, spatial cognition, engagement, and social influence were among the most reported mechanisms. Our results suggest that immersive technologies function as decision-shaping systems that alter how users perceive uncertainty, risks, consequences, and alternatives, highlighting the need for theory-driven research and evaluation in high-stakes decision contexts.

Computers doi: 10.3390/computers15070420

Authors: Yousef Sanjalawe Salam Al-E’mari Sharif Naser Makhadmeh

Advanced cyber threats increasingly exploit volatile memory to execute malicious payloads without touching persistent storage, rendering traditional disk-centric forensic tools insufficient for comprehensive digital investigations. This paper presents Forensic-BERT, an AI-driven forensic framework that automatically extracts and classifies potentially relevant artifacts from unstructured memory dumps across heterogeneous operating environments. The framework combines byte-boundary-preserving Hex-to-ASCII conversion, sliding-window Shannon entropy filtering (H>7.2 bits per byte, 256-byte windows) to isolate high-probability artifact regions, and a binary-aware WordPiece tokenizer extended with 2048 domain-specific tokens covering hexadecimal byte patterns, Windows API names, and Linux system-call sequences. These components feed a transformer-based classifier fine-tuned from bert-base-uncased (110 M parameters) on memory-derived text, with sliding-window inference and majority-vote aggregation for large images. A SHAP DeepExplainer module and averaged 12-head attention heatmaps provide transparent, analyst-accessible explanations for classification decisions. We evaluate the framework on a multi-source corpus of 735 labeled memory segments drawn from 197 distinct images across four independent collections, MemLabs, the DARPA Transparent Computing program, Digital Corpora, and live sandbox execution traces from Any.run and Joe Sandbox, spanning Windows XP through Windows 11, Ubuntu Linux 16.04/18.04, and FreeBSD. Source-stratified five-fold cross-validation yields an overall F1-score of 0.92±0.02 and AUC-ROC of 0.95±0.01 (95% CI). Forensic-BERT outperforms all six baselines, Volatility with YARA rules (F1 =0.71), Random Forest (F1 =0.82), BiLSTM with GloVe embeddings (F1 =0.85), MRm-DLDet (F1 =0.87), SPECTRE (F1 =0.89), and SecBERT (F1 =0.90), with every pairwise difference statistically significant under the McNemar test with Bonferroni correction. Explainability quality is independently confirmed by a Spearman rank correlation of ρ=0.81 between model SHAP token rankings and expert forensic-indicator rankings and by a System Usability Scale score of 73.2 among certified examiners. The complete pipeline processes 512 MB memory images in 7.5–10.2 s (GPU) or 38–52 s (CPU-only), scaling to 4 GB images with near-linear throughput. These results indicate that, on the corpus evaluated here, combining domain-adapted NLP preprocessing, transformer-based sequence modeling, and quantified explainability can improve the effectiveness and usability of analyst decision support and investigative triage for volatile memory analysis.

Computers doi: 10.3390/computers15070419

Authors: Mohammed Al Masarweh Tariq Alwada’n Adel Mohammad Hamdan Omar Almomani Isra’a Mustafa

High-performance resource orchestration and robust data security represent critical, often competing, operational objectives in modern cloud computing architectures. This study presents a unified infrastructure framework designed to reconcile these requirements by integrating a Geographically Aware Placement Algorithm (GAPA) with an Artificial Intelligence-Driven Monitoring system (AIDAM). GAPA dynamically optimizes workload distribution based on regional server capacity and geographic proximity, while AIDAM leverages deep unsupervised autoencoders for real-time anomaly detection and threat mitigation. The framework was evaluated via deterministic simulation using production traces from the Google Cluster Data (2019) corpus under a systematic injection of volumetric Distributed Denial-of-Service (DDoS) anomalies. The empirical results demonstrate a 92% macro-averaged threat detection accuracy rate against low-and-slow traffic variations alongside a minimal cryptographic processing latency overhead of 3–5% relative to an unencrypted baseline scheduling configuration. Furthermore, the integrated pipeline achieved a 25% reduction in end-to-end network latency compared to traditional non-geographically aware heuristic models. These findings demonstrate that cloud infrastructure efficiency and security resilience can be simultaneously enhanced without requiring comprehensive physical re-engineering.

Computers doi: 10.3390/computers15070418

Authors: Md. Arafatur Rahman Mohamad Ibrahim Bashir Ahmed Nadia Refat Tan Sze Wei Prashant Pillai

Human factors remain the dominant contributor to cybersecurity incidents, yet awareness training produces only moderate and often non-durable behaviour change, and most evaluated programs are either purely digital or evaluated only at the framework level. This study addresses two gaps: the scarcity of empirical and demographically stratified evidence for multi-modal community-facing awareness programs, and the lack of an explicit account of how artificial intelligence (AI) should be integrated into such programs rather than treated as an optional add-on. We evaluate Cyber4Me, a four-stage individual-awareness intervention (community roadshows, structured training, a hackathon, and a physical–digital escape room) that is wrapped in a cross-cutting AI adaptive layer built entirely on structured performance and behaviour data baseline competency tiering, awareness–behaviour gap detection, predictive early-warning, and personalised recommendation, with no reliance on free text. Using a single-group pre–post design with 130 participants in the UK Black Country region and a multi-dimensional Likert instrument, all four competency domains (confidence, familiarity, GDPR knowledge, incident-response preparedness) improved significantly (paired-t, all p<0.001; large within-participant effects, Cohen’s d≥1.0). Improvement was strongly moderated by demographics: older adults gained most in familiarity, undergraduates in confidence, and lower-education participants in regulatory knowledge. The contributions are as follows: transparent and demographically stratified pre–post evidence for a multi-modal awareness program with effect sizes reported; a fitness-for-purpose comparison against contemporary analogs (KnowBe4, Proofpoint, CyberPatriot, iCAT, CAT-RWE, GPT-CSAT, escape-room studies) that treats AI as a first-class design dimension; and an articulated AI integration architecture for the framework, demonstrated offline on the cohort using only structured performance and behaviour data (no free text). In this architecture, a gradient-boosted classifier assigns participants to three baseline competency tiers at 93.1% cross-validated accuracy; these tiers differ sharply in measured improvement (ANOVA F=68.8, p<0.001; Foundational +1.79 vs. Applied +0.30 scale points), an awareness–behaviour gap segment is detected and predicted from intake signals alone (AUC =0.73), and a recommender routes participants to personalised follow-on tracks. As the design is single-group and self-reported, results are reported as evidence of within-participant change associated with the intervention rather than as a causal efficacy estimate, and the AI layer is demonstrated for feasibility rather than being evaluated as a separate trial arm; the scope is explicitly individual security awareness and behaviour, not technical network, IIoT, or cloud security.

Computers doi: 10.3390/computers15070417

Authors: Rana Alrayes Atta Rahman

Materials such as steel, concrete, and various alloys are used to build infrastructure and machinery across all industries. Due to their long service life, some of these materials will eventually develop surface damage (such as crazing, corrosion, and pitting) that will negatively affect both the structural integrity and the reliability of the machinery/infrastructure. Thus, the rapid and accurate classification of defects on material surfaces is crucial for ensuring high-quality materials and a continuous process without machinery breakdowns. In this work, we compare the effectiveness of two types of deep learning models (a VGG16 convolutional neural network with transfer learning and the state-of-the-art YOLOv8) for automatic defect classification on surfaces. The dataset used in our experiment included data from the Phase 5 Capstone Corrosion and the NEU Surface Defects Databases, resulting in eight distinct classes of surface defects. The effectiveness of both models was determined using stratified 10-fold cross-validation. The results of the experiment revealed that YOLOv8 achieved 98.5% accuracy, whereas VGG16 achieved only 92.5%. Moreover, YOLOv8 exhibited greater consistency under noise perturbations, demonstrating superior robustness compared with VGG16. Beyond model comparison, this study introduces a unified benchmark constructed from heterogeneous industrial defect datasets. It systematically evaluates classification performance, generalization capability, and robustness using stratified cross-validation and noise-based testing. The results indicate that YOLOv8 is a practical solution for automated industrial surface defect classification.

Computers doi: 10.3390/computers15070416

Authors: Qunbo Wang Shiyi Xiong Jiawei Li Weiliang Li Chu Huang Sen Zhang Xize Guo Chao Fan Wenjun Wu

Railway scene understanding is critical for ensuring train operational safety and advancing intelligent railway systems. Existing railway vision methods mainly focus on perception and classification, while lacking regulation-guided semantic reasoning capabilities in complex environments. To address these limitations, this paper proposes a retrieval-augmented generation (RAG)-enhanced vision–language framework for railway signal cognition and safety reasoning. The proposed method integrates railway signal perception, regulatory knowledge retrieval, and multi-modal reasoning to improve factual consistency, reasoning reliability, and operational interpretability. In addition, a dedicated railway signal dataset comprising 500 standardized railway scene images with structured QA annotations is constructed to support regulation-oriented multi-modal recognition evaluation. Experimental results show that the proposed framework improves reasoning accuracy from 28.40% to 67.20% with an average end-to-end inference latency of 11.31 s per sample, and the inference speed can be further improved by adjusting experimental configurations to trade off between efficiency and accuracy, demonstrating the potential of RAG-enhanced architectures as a foundational step toward reliable multi-modal cognition in intelligent railway systems.

Computers doi: 10.3390/computers15070415

Authors: Dianjun Xie Wenai Song Ruize Guo Biaokai Zhu Yiran Li

Few-shot node classification aims to effectively predict new class nodes using only a small number of labeled samples, which is an important research direction in graph data mining. Existing self-training-based few-shot node classification methods are constrained by the bias and local optima of a single-teacher model. Meanwhile, multiple-teacher models often suffer from knowledge conflicts and redundant information, which degrade distillation efficiency and model generalization performance. To address these issues, we propose a multi-teacher knowledge distillation method with reinforcement learning for few-shot node classification (MTKD-RL). This framework is composed of a multi-teacher distillation network and a multi-teacher weight optimization module to deliver complementary supervision information from multiple perspectives. The reinforcement learning agent dynamically assigns adaptive weights to different teachers based on their prediction performance and the discrepancy between teacher and student models, which greatly enhances pseudo-label quality and distillation performance. Experiments on nine graph network datasets demonstrate that our method achieves consistent accuracy improvements ranging from 1.4% to 6.2% in few-shot node classification.

Computers doi: 10.3390/computers15070414

Authors: Yeison Alberto Garcés-Gómez Jhon Edwin Arias-Reyes Ángela Inés Guzman-Alvis Iván Felipe Benavides-Martínez Justin Guthrie

Mangrove ecosystems are critical for climate change mitigation; however, monitoring these environments in high-precipitation regions, such as the Colombian Pacific coast, is often hindered by persistent cloud cover and complex terrain. This study addresses these challenges by implementing the novel Google AlphaEarth Foundations (AEF) technology, leveraging 64-dimensional embeddings integrated with Digital Elevation Models (DEM) and Slope data. For classification, a Random Forest (RF) algorithm was deployed using a subset of only 7 embedding dimensions alongside topographical variables. The model estimated a mangrove extent of 209,262 ha, compared to a reference baseline of 137,732 ha. This discrepancy is hypothesized to stem from the model’s ability to map changes in the mangrove forest due to anthropogenic and natural factors and species migrating upstream, areas frequently overlooked in traditional inventories. The classification performance, evaluated on a spatially independent hold-out validation set, yielded an Overall Accuracy of 0.9844 and a Kappa Index of 0.9747. Regarding biomass estimation, the RF algorithm utilized 4 embedding dimensions plus DEM and Slope to achieve a Coefficient of Determination (R2) of 0.5844, a Root Mean Square Error (RMSE) of 18.99 Mg/ha, and a Mean Absolute Error (MAE) of 14.88 Mg/ha within a range of 100–300 Mg/ha. These metrics represent a notable advancement, successfully mitigating the physical signal saturation that typically constrains traditional single-sensor remote sensing models at high biomass thresholds. Significant advantages of this methodology include the complete elimination of cloud interference and a drastic reduction in processing time. These findings demonstrate that the synergy between foundational models and machine learning provides a robust, scalable, and efficient framework for managing blue carbon stocks in critical tropical regions.

Computers doi: 10.3390/computers15070413

Authors: Dianjun Xie Wenai Song Biaokai Zhu Ruize Guo Yiran Li

As the core automated execution components of blockchain technology, smart contracts enable programmatic control over digital assets; however, their immutable characteristics and inherent logical vulnerabilities give rise to substantial security risks. Although smart contract vulnerability detection methods based on large language models (LLMs) have exhibited certain potential in vulnerability detection and explanation, the coarse-grained modeling of traditional binary preference optimization paradigms hinders the model ability to learn the priority of domain-specific requirements, frequently leading to extreme optimization at the cost of detection accuracy. Furthermore, existing approaches fail to consider non-ideal factors in real-world application scenarios and overlook noise interference induced by missing prompts, which results in inadequate detection stability and reliability, making them challenging to adapt to complex practical scenarios. To address these critical issues, this study proposes a Noise-Aware Preference Reinforcement Large Language Model for Smart Contract Vulnerability Detection (NAPO-SCVD). This method adopts a four-stage framework consisting of data construction, continuous pre-training, supervised fine-tuning, and noise-aware preference optimization. Specifically, it enhances the model’s comprehension of contract syntax and semantics through domain-specific pre-training, improves its detection and explanation capabilities using high-quality datasets, constructs deliberately guided biased explanations to simulate noisy samples, refines preference gradients, and strengthens the model’s anti-interference ability. Consequently, this approach achieves high-precision and high-reliability smart contract vulnerability detection, along with fine-grained explanations.

Computers doi: 10.3390/computers15070412

Authors: Haodong Zou Yichen Zhao Xin Chen Ling Wang Jinghang Yu Long Yuan Luokai Jiang

In modern cloud-native systems, a single root cause can trigger cascading anomalies across multiple entities (e.g., microservices, databases, and hosts), generating alert storms with hundreds or thousands of heterogeneous alerts. Alert convergence (automatically grouping these alerts into actionable incident tickets) is critical for reducing operator burden and recovery time. Existing graph-based methods construct a topological graph from known entity dependencies and then leverage Graph Neural Networks (GNNs) for information propagation, but they rely on static physical topologies that fail to capture implicit fault propagation paths. Large Language Model (LLM)-based methods focus on reasoning about the textual information of alerts, yet they do not incorporate global topological structure and struggle with consistency at scale. Motivated by these limitations, we propose LLM-Guided Graph Structure Learning (LLM-GSL), a novel framework that combines the semantic reasoning ability of LLMs with the structural modeling power of GNNs for alert convergence. Specifically, LLM-GSL first leverages an LLM to evaluate pairwise entity relationships and discover implicit fault propagation paths that are absent from static topologies, thereby enhancing the physical-topology graph into a more complete structure. A Graph Attention Network (GAT) then refines alert representations over this enhanced graph via graph message passing, guided by a self-supervised graph affinity loss with continuous multi-modal supervision targets that fuse adjacency structure, textual affinity, and temporal affinity. Finally, density-based clustering groups the learned representations into incident tickets. Experiments on five public datasets, including four LogHub-derived datasets and one RCAEval microservice fault-injection subset, demonstrate that LLM-GSL achieves an average F1-score of 96.2%, outperforming six baselines including both traditional clustering and LLM-based methods by at least 14.0 percentage points.

Computers doi: 10.3390/computers15070411

Authors: Songshui Wu Guangyong Zheng Tao Jiang Jinke Yang

Few-shot font generation can be viewed as a challenging conditional image generation task, where the goal is to synthesize target glyphs from only a few reference samples while preserving structural fidelity and style consistency. This problem becomes particularly difficult for characters with complex spatial layouts and fine-grained stroke topology, where existing methods often struggle to simultaneously maintain structural integrity, local continuity, and stylistic coherence under sparse-reference conditions. To address this issue, we propose a skeleton-aware deformable alignment framework for few-shot font generation. Specifically, explicit skeleton priors are introduced into the diffusion-based generation process to provide structural supervision during denoising. In addition, a structure-constrained deformable content alignment module is designed to improve local feature correspondence while suppressing unreasonable geometric deformation. We further develop a multi-module content aggregation strategy to jointly model global layout patterns and local stroke details through complementary multi-level representations. Extensive experiments demonstrate that the proposed method consistently outperforms state-of-the-art approaches in both quantitative and qualitative evaluations. The results show that our method provides stronger structural preservation, better perceptual quality, and improved generalization in structurally complex glyph generation and cross-lingual style transfer.

Computers doi: 10.3390/computers15070410

Authors: Duc Nghia Vu Janos Demetrovics Hoang Son Nguyen

This paper addresses the critical challenge of identifying cost-effective coalitions of social media influencers to promote environmental sustainability (ES) messages under budget constraints. Traditional influencer marketing often relies on heuristics that ignore relational dependencies and heterogeneous agent costs, leading to redundant coverage and suboptimal resource allocation. To overcome these limitations, we introduce a novel Relational Cooperative Game (RG) framework that formalizes pre-determined dependencies among influencers and followers using closure operators, enabling a portfolio of polynomial-time identification algorithms for Minimal Winning Coalitions (MWCs). We further extend this model to the Weighted Cost-Relational Game (WCRG) to optimize campaigns with heterogeneous influencer costs. We prove that finding a Minimum-Cost Winning Coalition (MCWC) is NP-hard via reduction from weighted set cover and propose two complementary algorithms: (1) a Greedy Cost–Benefit (GCB) algorithm that operates in polynomial time and empirically achieves optimal solutions across all tested instances; a logarithmic approximation guarantee is established for the restricted single-antecedent model; and (2) an Integer Linear Programming (ILP) formulation enhanced with Strongly Connected Component (SCC) preprocessing to handle cyclic dependencies and yield exact optimal solutions for moderate instances. Extensive empirical validation, ranging from a representative six-agent cyclic scenario to large-scale synthetic networks (up to 300 agents), confirms the framework’s robustness and scalability. Results demonstrate that GCB consistently achieves optimal solutions (approximation ratio = 1.000×) with subsecond runtime (<0.2 s) and minimal memory overhead (<50 MB), while ILP-SCC leverages graph condensation for rapid exact solving. Compared to size-based baselines, WCRG achieves up to 95.2% cost savings by systematically leveraging cost-efficient micro-influencers, empirically validating that minimizing coalition size does not guarantee cost efficiency. These findings establish WCRG as a scalable, budget-aware optimization toolkit for maximizing the impact of sustainability campaigns through relational coalition design.

Computers doi: 10.3390/computers15070409

Authors: Aljazi Almarri Asayil Alawadh Maha Alsayed Muhammad Aasim Rafique

Enterprises are more inclined towards cloud adaptation for services including cloud databases, but there is persistent concern about data security. One of the major data security vulnerabilities arises from misconfigurations in authorization and authentication mechanisms of data management system. Identity and Access Management (IAM) serves as a general framework for security configuration in cloud database services, yet its manual and static policy administration results in misconfigurations, which include excessive rights, policy drift, illegal access, and so on. Thus, comprehensive reporting and a deeper understanding of the efficacy of alternative IAM security procedures in reducing security risks are needed. This study adopts a systematic process that looks at common IAM misconfiguration patterns in cloud database systems and examines relevant studies published between 2016 and 2026 to assess the effectiveness of automated IAM configurations. The literature advocates that manual IAM configurations significantly contribute to privilege escalation and security violations in cloud database services, whereas automated IAM approaches provide stronger protection through continuous enforcement, real-time monitoring, and improved visibility into access behaviors. Furthermore, the study identifies critical research gaps in real-time remediation and DBaaS-aware automation, which are necessary to mitigate security risks faced by enterprises, reduce IAM-related vulnerabilities, and enhance confidence in adopting cloud database services.

Computers doi: 10.3390/computers15070408

Authors: Durgesh Lohar Bibhash Sen Anupam Basu Seyed-Sajad Ahmadpour

Human-computer interaction (HCI) exploits various input methods to improve user experience, but those without visual access suffer more than the mainstream. In this context, this paper proposes a novel Gesture-based Human-Computer Interaction (GCI) system for visually impaired people (VIP). However, a large set of gestures introduces complexity, which poses challenges for VIP to interact with computers. Therefore, an accessible assistive application with a minimal set of gestures is designed here. Nineteen (19) participants engaged, and several dimensions were evaluated, including skin conductance, NASA-TLX, and performance indicators. The gesture response time revealed that the proposed GCI technique is 39% faster than the existing technique. In addition, the skin conductance revealed a modest reduction, which means GCI caused a more relaxed reaction than the existing technique. GCI demonstrated significant statistical advantages in gesture response time, skin conductance, and forgot word count, while other measures showed comparable performance between the two techniques. GCI provides a more efficient and cognitively favorable interaction experience which opens a new era in the design and development of assistive technologies for VIP.

Computers doi: 10.3390/computers15070407

Authors: Erkun Zhang Yu Zhang Pengxiang Xu Lu Lu

It is widely acknowledged that General Matrix Multiplication (GEMM) serves as a foundational kernel across numerous application domains. Complex numbers exhibit distinctive mathematical properties that enable their widespread adoption across engineering computing scenarios, including signal processing and signal transformation. This study investigates high-efficiency CGEMM, namely, complex-valued GEMM, for NPU hardware, broadening the application scope of NPUs beyond mainstream low-precision AI computation workloads. The major contributions of this study are as follows: (i) numerical precision and hardware utilization of the 3M and 4M decomposition schemes on Ascend NPUs are analyzed, and the 4M method is selected as the preferred CGEMM implementation under our tested hardware constraints to fit the bandwidth limitations of modern accelerators for both precision-sensitive and performance-critical matrix computation scenarios; (ii) a complete high-performance CGEMM design based on the 4M scheme tailored for Ascend NPUs is proposed, with an AIC/AIV dual-stream pipeline scheduling strategy equipped to coordinate padding operations, matrix–matrix multiplications, and element-wise instructions across multi-level memory hierarchies and compute units; (iii) a fine-grained task scheduling and assignment mechanism is implemented to maximize Cube core occupancy across diverse matrix dimensions, improving hardware utilization for various computation workloads. Our experimental measurements show that the proposed CGEMM achieves a competitive hardware utilization rate of 83.6% across all tested matrix configurations, enabling efficient exploitation of available computing resources. Meanwhile, we observe a measured average speedup of 1.14× relative to the AscendSipBoost implementation tested on an identical Ascend NPU, alongside a measured 3.17× speedup compared with cuBLAS running on the Nvidia GPU platform adopted in our experiments across all evaluated matrix sizes. These results reflect the promising capability of Ascend NPUs for high-precision complex-valued computing workloads within the tested experimental setup.

Computers doi: 10.3390/computers15070406

Authors: Jianan Li Xiaonan Chai Wei Gao

While the Triton compiler has revolutionized GPU kernel development, its deployment on general-purpose CPUs struggles to fully utilize the underlying hardware capabilities. This is primarily due to the semantic gap between Triton’s SPMD execution model and CPU vector architectures, which leads to suboptimal utilization of vector units during complex memory accesses. In this paper, we present a comprehensive compilation and runtime co-optimization framework for Triton-CPU, specifically targeting Vector Length Agnostic architectures (VLA) like ARM SVE. At the compiler level, we propose a novel semantic reconstruction and explicit base-offset decoupling strategy, enabling native VLA gather/scatter generation and eliminating scalar loop overheads. At the runtime level, we introduce a Machine Learning-driven thread scheduling model to optimally orchestrate the synergy between Thread-Level Parallelism and Vector-Level Parallelism. Extensive evaluations on an ARM-based multi-core processor demonstrate that our framework achieves up to a 2.0× throughput improvement for compute-bound GEMM operators (peaking at 346 GFLOPS), notably outperforming the hand-optimized OpenBLAS library by up to 1.54× at small-to-medium scales. Additionally, it delivers a 1.7× speedup for element-wise workloads. Furthermore, our optimizations saturate memory bandwidth (up to 55 GB/s) for memory-bound operators with zero compilation bloat, establishing a robust, high-performance foundation for deploying deep learning models on general-purpose CPUs.

Computers doi: 10.3390/computers15070405

Authors: Nicolas Afonso-Alonso Juan A. Holgado-Terriza Miguel A. Oltra-Rodríguez Paul Stonehouse

Agile teams often struggle to translate business requirements into maintainable, high-quality software due to the persistent ambiguity in the roles and relationships of behavior-driven development (BDD), Acceptance Test-driven Development (ATDD), and Test-driven Development (TDD). These approaches are frequently misunderstood, inconsistently applied, and only loosely connected within a unified delivery lifecycle. This article introduces BLOW (Behavior-Driven Development in a Layered Organization of Work-Centers), a systematic approach that establishes BDD as the coordinating methodology between ATDD (business-focused) and TDD (technology-focused). BLOW structures scenario-driven development across layered domains of accountability with clearly defined roles and responsibilities, organizing delivery through nested work-centers that transform user stories into executable specifications and production code. This approach integrates two complementary collaboration practices: the Three Amigos for discovering and formulating business scenarios, and the proposed Technical Three Amigos for linking those scenarios to Technical Domain Contexts, identifying required Enablers, and deriving technical scenarios when additional architectural support is needed. The proposed operating model emphasizes observability through executable scenarios as first-class artifacts, introducing native, test-anchored metrics that support reasoning about progress, technical effort, and value delivery within scenario-driven development. An exploratory longitudinal case study, consisting of a single-sprint proof of concept followed by an 18-month production deployment, reports patterns in which technical enablement precedes business value delivery and reusable infrastructure supports sustained growth of business scenarios over time. The findings also indicate that changes in the applied operating model are associated with measurable shifts in scenario evolution and internal quality indicators. Overall, BLOW provides a governance-compatible, end-to-end approach for organizing scenario driven development and improving alignment between stakeholder intent and technical implementation in complex software systems.

Computers doi: 10.3390/computers15070404

Authors: Bálint Dominik Orosz Judit Szücs Máté Cserép

Static analysis of student programming submissions has proven a useful supplement to manual evaluation in university courses, but existing approaches focus on local code-quality issues and rarely check higher-level design decisions such as architectural conformance. We propose an architecture-aware static-analysis methodology for student submissions written in C# and structured according to the Model-View (MV) or Model-View-ViewModel (MVVM) architectures. A deterministic clustering algorithm assigns user-defined types to architectural layers by combining heuristic rules derived from SDK conventions with course-specific information, and our 10 proposed violation checks—covering layer-dependency rules, encapsulation, event handling, and dependence on concretions—are evaluated on the recovered layer structure. We implemented the methodology as an open-source analyzer integrated with an automated submission-evaluation system used in a university course focused on event-driven applications, and evaluated it on 947 submissions containing 13,126 user-defined types from past semesters. The analyzer assigned more than 98% of types to their correct layer and surfaced more than 6000 architectural and design issues. The results show that architecture-aware static analysis is a viable complement to manual grading and produces actionable feedback for both students and lecturers.

Computers doi: 10.3390/computers15070403

Authors: Mohammed Chaman Mohamed Benaly Anas El Maliki Wiame Bouyoussef Azzedine El Mrabet Hamad Dahou Abdelkader Hadjoudja

The increasing demand for intelligent transportation systems (ITS) and advanced driver assistance system (ADAS) significantly demands a real-time and robust perception to recognize road-side obstacles in varying different weather settings. This paper presents DriveEdgeAI, a lightweight YOLO11 based embedded deep learning framework for efficient road anomaly detection with the emphasis on potholes, speed bumps and relevant traffic sign detection. We have prepared a custom dataset consisting of 17,061 annotated images to train and test the model under different lighting conditions, weather conditions, and roads configurations. The proposed system also managed to demonstrate good convergence and generalization with a precision@50 of 95.8%, recall@50 of 89.7%, mAP@50 of 95.4%, surpassing previous YOLO versions. The stability and robustness of the model at different thresholds were also substantiated by Precision-Recall and F1-Confidence analyses. DriveEdgeAI was also deployed on a number of edge devices, such as Jetson Nano, Raspberry Pi 5, Intel Movidius VPU and Hailo-8L NPU respectively reaching 9.5 FPS/W and 28.5 FPS for the Raspberry Pi 5 + Hailo-8L version. From these results, one can conclude that DriveEdgeAI is an energy-efficient and scalable solution for real-world ADAS applications.

Computers doi: 10.3390/computers15070402

Authors: Bigul Mukhametzhanova Akgul Naizagarayeva Gulbakyt Ansabekova Shynar Turmaganbetova Yermek Sarsikeyev Akmaral Kassymova Azamat Dnekeshev Pavel Dunayev Zhanat Manbetova

This study presents a hybrid artificial intelligence system for latent cardiovascular risk stratification based on publicly available clinical and laboratory data. The proposed system integrates data preprocessing, auxiliary target modeling, latent phenotyping using UMAP and Gaussian mixture models, fuzzy logic-based risk integration, and multilevel predictive modeling. The key contribution of the system is the construction of a proxy target reflecting latent risk progression by combining phenotypic structure, probabilistic indicators, and mortality-related anchor points. Experimental evaluation was conducted on the NHANES dataset. The final analytical cohort included 78,822 adult participants, and the modeling set was divided into training, validation, and test subgroups using a stratified 70/15/15 design. The proposed PhaseFuzzy Hybrid model achieved an accuracy of 0.8390, a balanced accuracy of 0.7302, an F1-score of 0.5225, an MCC of 0.4203, an ROC-AUC of 0.8489, a PR-AUC of 0.5014, and a best LogLoss value of 0.4290 on the test set. The latent phenotyping step also demonstrated acceptable internal validity with a silhouette coefficient of 0.4138 and a confidence of 0.8800. The results demonstrate that the proposed framework identifies hidden cardiometabolic risk factors and provides an interpretable, scalable, and calibration-aware framework for latent cardiometabolic risk stratification and population-level screening.

Computers doi: 10.3390/computers15070401

Authors: Anita Jansone Zanda Aivita Cīrule

This study examines the development and pedagogical impact of an AI-based pedagogical agent designed for modern e-learning environments. The research addresses a key challenge in digital education: the lack of personalization and immediate feedback in traditional e-learning systems. AI-driven agents “support and motivate learners through instructional interaction” and provide adaptive, data-driven learning experiences that surpass the limitations of rule-based systems. The study begins with a systematic literature review following PRISMA 2020, analyzing 46 publications from 2020 to 2025 to identify current AI architectures, pedagogical roles, and the empirical evidence of learning impact. The findings highlight the growing use of machine learning, deep learning, multimodal analytics, and large language models in educational agents. These systems perform roles such as tutor, coach, evaluator, dialogue partner, and consultant, offering cognitive, metacognitive, emotional, and analytical support. Modern agents “continuously monitor user interaction, analyze engagement, and adapt learning content”, enabling highly personalized learning pathways. The study also presents the design of a multimodal pedagogical agent capable of explanation, task generation, diagnostics, and adaptive feedback. Experimental results with students (n = 20) show improved performance, reduced errors, and higher engagement when learning with the agent. Overall, the research demonstrates that AI-based pedagogical agents enhance learning effectiveness and support autonomous learning in higher education.

Computers doi: 10.3390/computers15070400

Authors: Md Eahia Ansari Md Tanvir Rahman Tarafder Abir Chowdhury Nur Nahar Rimi Nipa Akter Khandakar Rabbi Ahmed

Employee attrition is a significant organizational challenge associated with substantial financial costs and the erosion of institutional knowledge. This study presents an AI-based Management Information System (MIS) that integrates machine learning (ML) models to forecast employee turnover and support technical interpretability for HR decision-making. Using the IBM HR Analytics Dataset comprising 1480 employee records with 38 features, we implemented a rigorous preprocessing pipeline—including Synthetic Minority Over-sampling Technique (SMOTE) applied exclusively within training folds to prevent data leakage, one-hot encoding, Z-score normalization, and mean-value imputation. Four ML classifiers—Logistic Regression (LR), Random Forest (RF), Multi-Layer Perceptron (MLP), and XGBoost—were evaluated under a stratified 80/20 split with 5-fold cross-validation. XGBoost achieved the highest performance, attaining an accuracy of 87.83%, a ROC-AUC of 0.94, a PR-AUC of 0.96, and an F1-score of 93.04%, attributed to its sequential boosting mechanism and built-in L1/L2 regularization. Beyond predictive performance, the system incorporates SHapley Additive exPlanations (SHAP) to deliver feature-level transparency, enabling HR professionals to engage in proactive, informed retention interventions while retaining full decision-making authority. Within-dataset comparisons confirm that the proposed framework outperforms prior methods evaluated on the same benchmark; cross-study accuracy comparisons are reported as contextual reference only, given differences in datasets and experimental protocols. The system facilitates human oversight by positioning AI as a decision-support collaborator rather than an autonomous replacement in workforce management. Future work will address real-time deployment, controlled user studies with HR practitioners, and validation with actual organizational HR data.

Computers doi: 10.3390/computers15060399

Authors: Ivan Torlakov Yuri Zhelyazkov

DoS and DDoS attacks remain a major threat to service availability in modern IP and IoT networks, yet many learning-based detectors depend on dataset-specific flow exports, feature tables, or preprocessing conventions. This article presents a unified sequence-level detection pipeline designed to process heterogeneous public datasets through the same representation. Raw PCAP/PCAPNG traces from CIC-IDS-2017, CIC-DDoS-2019, and CICIoT2023 are converted into one-second aggregates per destination host using header-only features derived from IP, TCP, UDP, and ICMP metadata, source diversity, and packet timing. Dataset-specific annotations are used only to assign binary DoS/DDoS labels to this common representation. The resulting time-ordered aggregates are grouped into fixed-length temporal windows and classified by a compact transformer encoder, TemporalDosTransformer, which produces a window-level attack probability. The study focuses on whether a clean PCAP-based aggregation and labelling flow can support consistent DoS/DDoS detection across multiple datasets without payload inspection, flow-exporter dependence, or dataset-specific feature engineering.

Computers doi: 10.3390/computers15060398

Authors: Valerio Mandarino Giuseppe Pappalardo Emiliano Tramontana

Authentication is essential to hold users accountable across online services. Conventional authentication systems rely on centralized architectures or third-party identity providers, which, however, introduce single points of failure, privacy concerns, and limited user autonomy. Conversely, fully decentralized authentication frameworks often struggle to provide reliable identity attestation mechanisms. This makes them vulnerable to Sybil attacks and self-asserted claims, while limiting their interoperability with trust-based systems. This paper presents dAuth, a hybrid blockchain-based authentication architecture based on Ethereum smart contracts to provide cryptographic tokens that enable authentication to services. These tokens, anchored to the smart contract, are derived by users from institutionally certified base credentials issued by an accredited verifying authority and enable authentication to services without further involvement of the authority. Each token is cryptographically bound to a specific service, constrained in scope and duration, and verifiable off-chain through data and cryptographic commitments provided by the user. No plaintext personal information is published on-chain: identity attributes are committed as cryptographic digests, which anchor certified identity data on-chain while keeping the underlying personal information private and auditable. This design removes the verifying authority from the authentication process, as all authentication steps are assisted by the user-controlled smart contract. The verifying authority’s role is limited to initial identity certification and exceptional update procedures. The result is a privacy-preserving and verifiable hybrid authentication framework that leverages the cryptographic security properties of the underlying blockchain infrastructure and inherits its scalability characteristics. The proposed design has been implemented and experimentally evaluated on the Ethereum platform, addressing public blockchain-specific challenges such as scalability constraints and transaction costs to ensure practical deployment.

Computers doi: 10.3390/computers15060397

Authors: Jianzhuo Li Panyu Dai Jiake Li Ye Yu

Driving behavior recognition plays a crucial role in intelligent driving systems and road traffic safety. Due to insufficient driving experience and limited ability to allocate visual attention, novice drivers are considered a high-risk group for traffic accidents. Existing approaches primarily focus on experienced drivers and rely on single-modal eye-tracking data, making it difficult to model spatial attention distributions and long-term temporal dependencies simultaneously. Moreover, these methods are often affected by modality asynchrony during multimodal fusion, further limiting performance gains. To address these challenges, this study proposes a novice driver behavior recognition method based on bimodal eye-tracking features and a gated cross-modal attention fusion (GCMAF) mechanism. The model adopts a spatial–temporal dual-branch architecture. The spatial branch employs ResNet34 to extract eye-tracking heatmap features to represent the visual attention distribution. In contrast, the temporal branch integrates a 1D-CNN with the Mamba model to capture local dynamic patterns and long-range temporal dependencies. In the fusion stage, the GCMAF module is introduced to enhance cross-modal interactions, and a gating mechanism is further used to adaptively adjust modality weights, thereby mitigating the adverse effects of modality asynchrony. To validate the effectiveness and generalization ability of the proposed method, repeated experiments and five-fold cross-validation are conducted. The results demonstrate that the model achieves an average classification accuracy of 93.86% across four driving behavior categories, with standard deviations below 0.3%. Compared with baseline methods, paired t-test results show that the performance improvement is statistically significant (p < 0.01). Ablation studies further confirm the independent contribution of each component. Overall, the proposed method outperforms existing approaches in terms of accuracy and stability, providing effective support for driving behavior assessment and proactive safety warning systems.

Computers doi: 10.3390/computers15060396

Authors: Lei Wang Yu Gao Xiaoyong Zhao

Power systems continuously generate large-scale load time series data for forecasting, consumption analysis, and equipment health monitoring. However, real-world load measurements are often contaminated by anomalies caused by sensor faults, communication errors, and abnormal consumption behaviors, which may degrade data quality and affect operational decision-making. To address this issue, this paper proposes an integrated anomaly detection framework named PEL, which combines Prophet-based seasonal-trend decomposition, ensemble empirical mode decomposition (EEMD), and a multilayer long short-term memory (LSTM) network. Prophet is first employed to decompose the original series into trend, seasonal, holiday, and residual components. Sample entropy analysis and white noise tests are then adopted to evaluate whether the residual component still contains complex structured information requiring secondary decomposition. Next, EEMD is applied to the residual component to extract multi-scale intrinsic mode functions. Finally, all decomposed components are normalized and fed into a multilayer LSTM model for anomaly detection. Experiments on a real-world power load dataset demonstrate that the proposed PEL framework achieves an accuracy of 99.92%, a precision of 97.33%, a recall of 100%, an F1-score of 98.65%, and an AUC of 0.9996, outperforming or matching several baseline and hybrid models.

Computers doi: 10.3390/computers15060395

Authors: Yuze Gao Jiamin Zhu Xiaoxiao Liu Wei Wu

Edge vision systems require convolutional neural networks (CNNs) that preserve recognition accuracy under strict storage, computation, and latency constraints. Although ResNet18 is a compact residual backbone, direct deployment on resource-constrained devices remains costly, whereas simple channel reduction weakens representation capacity. This study aims to build a deployable ResNet18-based classifier that reduces model complexity while recovering the accuracy lost during compression. We propose a lightweight framework that combines global channel scaling, a re-parameterized attention residual block, and teacher–student knowledge distillation. The proposed block uses multi-branch convolution and squeeze-and-excitation attention during training, then folds the linear branches into a single 3-by-3 convolution for inference. Experiments on CIFAR-100 show that the final model reduces parameters from 11.220 M to 2.841 M, retains comparable Top-1 accuracy (0.7579 vs. 0.7606), improves Top-5 accuracy (0.9340 vs. 0.9253), and reduces graphics processing unit (GPU) batch inference latency from 3.279 ms to 2.161 ms. Deployment on PYNQ-Z2 verifies the complete camera-based CPU-side inference workflow, with an average end-to-end latency of 421.467 ms/frame. The results indicate that residual topology preservation, re-parameterized feature enhancement, and distillation form a practical route for edge-oriented lightweight CNN deployment.

Computers doi: 10.3390/computers15060394

Authors: Libero Nigro Franco Cicirelli

The work described in this paper continues basic research aimed at improving clustering algorithms such as K-Means and Random Swap through careful seeding and genetic concepts. This paper, in particular, develops a variation in the Hartigan–Wong (HW) algorithm, which, although computationally more expensive, is recognized as a better solution than K-Means. The new algorithm is named Recombinator Hartigan–Wong (Rec-HW). Rec-HW first builds a population of candidate solutions, each tailored to the minimization of the Sum-of-Squared-Errors (SSE) objective function cost. Candidate solutions are then systematically recombined by exploiting the standard behaviour of HW, which performs crossover and mutation operations. Recombinations, as experimentally confirmed, reduce the number of iterations required by basic HW and tend to favour the emergence of a solution close to the optimal one. The paper describes the design of Rec-HW, whose current implementation depends on parallel Java. Good clustering performance is demonstrated by using both benchmark and real-world datasets.

Computers doi: 10.3390/computers15060393

Authors: Dan Zhang Thuong Hoang Ye Zhu Rui Wang Paula Crouch Yi Wang

Generative large language models (LLMs) are increasingly used to support writing feedback. However, the pedagogical safety and usefulness of LLM feedback for primary students remains under-evaluated. This study reports an educator-centered evaluation of GPT-4 Turbo for Year 5 narrative and persuasive writing in the context of an established online tutoring program. Using authentic students’ drafts paired with tutor feedback, we generated parallel LLM feedback via rubric-aligned prompting and compared the two feedback sources in a blinded, within-script design. Four experienced English specialists co-designed a six-dimensional rubric (clarity, specificity, helpfulness, feasibility, relevance, and overall effectiveness) and rated tutor versus LLM feedback for each script; their written reflections were analyzed thematically to surface boundary conditions and risk perceptions. Across dimensions, tutor feedback received slightly higher mean ratings, with the clearest descriptive advantage in perceived helpfulness; however, none of the differences remained statistically significant after Holm-Bonferroni correction. LLM feedback was often rated similarly for clarity and feasibility but was frequently characterized as generic, surface-focused, and occasionally misaligned with the student draft, which increased verification effort and posed a risk of misleading learners if used without mediation. Synthesizing ratings and educator reflections, we identify conditions under which LLM feedback is most appropriate as rapid first-pass support for routine structure and surface revision, and least appropriate for developmental judgment and context-sensitive guidance. We translate these findings into design requirements for teacher-in-the-loop primary writing feedback systems, including alignment to explicit pedagogical constructs, editable workflows, and safeguards that reduce unsupported feedback before release to students.

Computers doi: 10.3390/computers15060392

Authors: Sarunya Kanjanawattana Isaman Sangbamrung Dulyawat Wiriyaphong Gun Bhakdisongkhram

An electroencephalogram (EEG) signals provide vital insights for stroke rehabilitation, yet analyzing these complex, high-dimensional data to detect gait anomalies remains challenging. Artificial intelligence offers a promising solution to precisely identify abnormal movements, assisting physicians in optimizing personalized treatments. This exploratory pilot study aims to evaluate multi-class deep learning frameworks for classifying eight distinct normal and abnormal motor activities in stroke patients using EEG data. EEG signals from eight stroke patients were utilized to train and evaluate a customized Convolutional Neural Network (CNN), DeepConvNet, and EEGNet. Furthermore, channel reduction configurations (32, 22, and 15 channels) were investigated to determine optimal clinical setups. In the Leave-One-Out Cross-Validation (LOOCV) evaluation involving seven patients, EEGNet attained the highest descriptive average F1-score of 0.810. Moreover, when assessed independently on an unseen patient, it achieved an F1-score of 0.915, indicating its potential in accommodating individual differences within this limited cohort. Moreover, EEGNet exhibited a low false positive rate of 0.175, minimizing false alarms. While the 32-channel setup yielded the highest consistency, reduced configurations served as hypothesis-generating for specific tasks. In conclusion, EEGNet demonstrated superior average performance in differentiating complicated gait patterns in this exploratory pilot study, underscoring its promise for real-time, non-invasive monitoring in stroke neurorehabilitation.

Computers doi: 10.3390/computers15060391

Authors: Sari Al-serri Mohammad Reza Chalak Qazani Shady Mohamed Saeid Nahavandi Houshyar Asadi

High-fidelity motion cueing in driving simulators is essential for delivering a realistic and immersive user experience. However, the trade-off between motion accuracy and computational efficiency often hinders achieving this. Fixed-horizon Model Predictive Control (MPC)-based Motion Cueing Algorithm (MCA) frameworks frequently struggle to adapt to rapid dynamic changes in vehicle behaviour, resulting in suboptimal simulator responses. Their reliance on worst-case horizon tuning can result in inefficient platform usage and increased computational load, limiting computational efficiency and practical deployment. This study presents an adaptive MPC-based MCA designed to enhance the fidelity of motion platforms used in vehicle dynamic simulations. The proposed method dynamically adjusts the MPC prediction horizon to improve overall simulation performance while minimising motion sensation error. Within the simulation environment, the prediction horizon is adaptively updated at each simulated control step according to recent tracking-performance metrics, enabling responsiveness to varying vehicle dynamic models and driving scenarios. The system was developed and implemented using Python and MATLAB environments, with Long Short-Term Memory (LSTM) networks employed to enhance the adaptability and precision of prediction horizon adjustments. Due to safety constraints, the proposed framework was evaluated exclusively within a simulation environment and compared against both classical MPC-based MCA and RL MPC-based MCA. Experimental results demonstrate that the proposed adaptive framework improves workspace utilisation and substantially reduces computational load compared with the classical and RL-based MPC-based MCA approaches, while maintaining competitive motion cueing tracking performance. The adaptive system effectively enhances linear displacement (LD), ensuring better alignment of motion cues with platform constraints. While minor trade-offs were observed in root mean square error (RMSE) and correlation coefficients (CCs) for sensed angular velocity (SAV) and sensed specific force (SSF), the framework improves workspace utilisation and computational efficiency while maintaining competitive motion cueing performance. Furthermore, the adaptive LSTM-MPC framework substantially reduces computational load, achieving approximately 44.26 times faster execution compared with the classical MPC-based MCA and approximately 30.03 times faster execution compared with the RL MPC-based MCA. These findings highlight the potential of integrating deep learning (DL) with MPC to optimise the trade-off between motion cueing performance, platform utilisation, and computational efficiency in driving simulators.

Computers doi: 10.3390/computers15060390

Authors: Zepeng Chen Hui Liu Lin Zhang Chenjie Wu

Existing cross-chain bridges over-rely on static collateralization and post-event penalties, leaving them vulnerable to concealed on–off attacks and rational group collusion. To address these limitations, this paper proposes a Dynamic Trust Evaluation and Risk Control (DTERC) mechanism for heterogeneous cross-chain relay nodes. First, DTERC develops a multidimensional trust quantification model that combines temporal decay, robust multi-observer latency aggregation, verification accuracy, online stability, and an asymmetric one-strike penalty triggered only by cryptographic evidence. Second, DTERC constructs a threshold-aware N-player evolutionary game model to characterize the k-of-N signature structure of cross-chain relay consensus and introduces a dynamic staking function to reduce the economic incentive for collusion under bounded attack-value and parameter conditions. Third, DTERC designs a threshold-preserving FastPath mechanism to reduce redundant verification for low-risk transactions while retaining committee-level confirmation and challenge-based fallback. The empirical evaluation combines multi-agent simulation, smart-contract prototype testing, whitelist-compromise stress tests, malicious-oracle robustness analysis, network-jitter experiments, repeated trials, and parameter-sensitivity analysis. The results show that, under the tested settings, DTERC reduces the malicious transaction success rate to 0.15% under a 50% initial collusion scenario, lowers core contract Gas overhead by 35.7%, and reduces average end-to-end latency by approximately 10% in benign FastPath conditions. These findings indicate that DTERC improves the security–efficiency trade-off of heterogeneous cross-chain relay networks while making its assumptions and limitations explicit.

Computers doi: 10.3390/computers15060389

Authors: Bandana Mallick Priyadarsan Parida Bibhu Prasad Chittaranjan Nayak Manoj Kumar Panda Nawaf Ali N. Mohan Kumar

Cyber-physical systems (CPSs) based on the Internet of Things (IoT) form the backbone of modern smart infrastructures, including smart cities, healthcare monitoring, industrial automation, and intelligent transportation. However, connecting many resource-limited IoT devices makes them more vulnerable to cyber threats, particularly quantum attacks. This review comprehensively examines quantum-secure communication (QSC) frameworks for IoT-enabled CPS, focusing on Quantum Key Distribution (QKD), post-quantum cryptographic (PQC) algorithms, and hybrid quantum–classical security models suitable for constrained devices. A PRISMA-guided search of the Scopus and Google Scholar database was conducted in January 2026 using three keyword groups related to hybrid security, artificial intelligence, and cyber-physical systems. Based on the evaluation, 6008 publications have been identified between 2001 and 2026. The first-round screening was performed for 4948 articles, after excluding duplicates. During the screening stage, 348 articles were selected for abstract scrutiny, 115 records were excluded due to no direct focus on CPS/IoT applications, 52 studies were excluded because these papers relied on traditional security models, 25 studies were excluded due to insufficient relevance to the review objectives, and 15 additional non-English studies were removed. Following the screening stage, 141 studies were selected for full-text eligibility. Out of those, 86 studies were removed due to a lack of specific evaluation metrics or not being published in a peer-reviewed venue. Furthermore, the publications are classified as QKD-based secure CPS and QSC for industrial IoT, AI-Assisted Secure Communication for CPS Networks, and hybrid PQC-QKD models for CPS/IoT devices. This article investigates recent advancements in secure data transmission, verified protocols, and AI-driven anomaly detection customized to CPS/IoT environments. In addition, operational hurdles, interaction with open innovations, real-time deployment, and secure edge-cloud integration are highlighted. By analyzing recent developments and identifying research gaps, this review provides a structured roadmap for designing secure, scalable, and quantum-safe IoT-based CPS frameworks capable of withstanding next-generation cyber threats. This systematic review was performed and reported according to the PRISMA 2020 guidelines.

Computers doi: 10.3390/computers15060388

Authors: Alibek Barlybayev Marek Milosz Nurzada Amangeldy Guohui Li Bibigul Razakhova Aruzhan Tazhibay Aizhan Nazyrova Zhanar Lamasheva

This systematic review investigates the application of computer-vision technologies for automated monitoring of personal protective equipment compliance in industrial environments. This review followed the PRISMA 2020 guidelines and covered studies published between 2010 and 24 February 2026. It provides a structured synthesis of advances in deep learning-based object detection models, with particular emphasis on different YOLO variants, two-stage detectors such as Faster R-CNN, and emerging transformer-based and vision–language models. Model effectiveness, reported performance metrics, and dataset characteristics are comparatively examined, including their performance under practical operating conditions. Special attention is given to performance variability in real-world scenarios affected by illumination changes, occlusion, viewing angle variation, worker movement, computational constraints, and large-scale deployment requirements. The review also appraises the reporting quality and risk of bias of the included studies and identifies current research trends, methodological limitations, and the gap between laboratory validation and industrial implementation. It also outlines future directions for improving the reliability, cost-effectiveness, and practical application of computer vision-based personal protective equipment compliance systems.

Computers doi: 10.3390/computers15060387

Authors: Qi Yang Jing Li Aoyun Zhu Hao Chen

Affected by inter-annotator cognitive differences, fatigue effects, and data poisoning, training data inevitably contains a certain proportion of noise, which severely impairs model performance. Traditional manual verification is costly and inefficient, while existing automatic detection methods generally suffer from limited precision, poor interpretability, and insufficient robustness. This paper proposes a noise label detection and correction method based on Bayesian weighted consensus inference. First, an ensemble of multiple lightweight heterogeneous models is constructed, and model prior knowledge and dataset noise are obtained on a clean validation set. Second, the model ensemble predicts noisy samples to extract two-dimensional consensus evidence. Then, prior knowledge and consensus evidence are fused, and the posterior probability of label noise is calculated via Bayesian inference to generate correction suggestions. Finally, high-confidence noisy labels are precisely screened based on the posterior probability threshold. Experimental results on three datasets show that the proposed method achieves a precision of 96.50%, a recall of 98.61%, an F1-score of 97.54%, and a correction accuracy of 95.53%, with improvements of 5–20% over mainstream methods. With a computational cost comparable to that of basic ensemble methods, the proposed approach achieves a favorable balance among precision, robustness, and interpretability. It thus offers a promising and cost-effective solution for automated quality control of large-scale annotated datasets, especially in text classification tasks.

Computers doi: 10.3390/computers15060386

Authors: Zhenyu Yang Biao Song Yuan Tian

The rapid growth of scientific spatiotemporal data poses increasing challenges for efficient storage and transmission while preserving sufficient reconstruction fidelity for downstream analysis. Existing compression methods remain limited in this setting: lossless approaches often yield low compression efficiency, conventional lossy methods lack flexible local fidelity control, and learning-based schemes may introduce oversmoothing in reconstructed results. To address these issues, we propose a hybrid machine-learning framework for error-bounded compression of gridded scientific data. The proposed framework integrates MPEG-based temporal coding and transformer-based super-resolution reconstruction to exploit temporal correlation and spatial redundancy, and it introduces an error-bounded correction module to explicitly control local reconstruction errors. In addition, a lightweight Dense Residual Swin Transformer (DRCT)-based reconstruction model is employed to enhance long-range dependency modeling and multi-scale feature recovery. Experimental results on Copernicus Marine Service (CMEMS) gridded sea-level data demonstrate that the proposed framework achieves a favorable balance between compression efficiency and reconstruction quality. For 192×192 ADT data, the method reaches a peak PSNR of 38.14 dB with a compression ratio of 594×. With the error-bounded correction module enabled, the reconstructed values are first de-normalized using the recorded normalization parameters, and the local reconstruction error in the original floating-point physical-value domain can then be explicitly controlled by the prescribed absolute error threshold while maintaining a compression ratio of 295×. Additional experiments on SLA further indicate that the proposed framework is not restricted to a single sea-level variable. These results indicate that the proposed framework is a practical and effective solution for compressing large-scale scientific spatiotemporal data with controllable error and high-fidelity reconstruction.

Computers doi: 10.3390/computers15060385

Authors: Răzvan Bologa Andrei Toma Corina-Marina Mirea Dimitrie-Daniel Plăcintă Aura Elena Grigorescu Iulian Întorsureanu Dragoș-Marcel Vespan Alina-Mihaela Ion Lorena Bătăgan Sergiu Costan

This article examines the medium to long-term enrollment patterns of an AI-based platform designed to support children in learning robotics and participating in a national robotics competition in Romania. Drawing on registration and participation data covering students and teachers across urban and rural schools between 2020 and 2025, the study documents a consistent pattern: an initial period of high enrollment and rapid adoption followed by a steady decline over time. A key feature of the initiative is that hardware, platform access, and learning resources were provided entirely free of charge, allowing cost-related explanations for the decline to be set aside and structural and human factors to be examined directly. The paper makes two primary contributions. First, it proposes a System Dynamics framework grounded in innovation diffusion theory as a first-generation calibration model for understanding AI-enabled educational robotics adoption in a resource-constrained national context. The model is designed to be progressively tested and refined as anonymized aggregate data accumulates, and it relies exclusively on anonymized aggregated public data in accordance with GDPR requirements. Second, it advances the hypothesis that an AI-based educational platform, even one from which all financial barriers have been removed, will experience sustained enrollment decline in the absence of adequate human teacher involvement. The empirical trajectory and model outputs are consistent with this hypothesis and motivate further investigation. This represents a hypothesis-generating and framework-building paper. The framework reveals pronounced urban-rural disparities and differential outcomes by age of entry. All findings are presented as model-generated hypotheses rather than empirically demonstrated conclusions. The paper invites researchers gathering comparable data from similar initiatives in other countries to collaborate in testing and refining the model. The central conclusion is cautiously optimistic: AI may support robotics education adoption, but it is not a substitute for dedicated teachers, and without sustained investment in human capital, even a financially accessible platform is insufficient to maintain long-term enrollments.

Computers doi: 10.3390/computers15060384

Authors: Raffaella Folgieri Claudio Lucchiari Sergej Gričar Tea Baldigara

Virtual Reality (VR) is increasingly used in clinical, educational, and supportive-care contexts, but evidence on VR-based creative interventions for vulnerable populations remains fragmented. This article presents a scoping review and proposes VR-CREAT (Virtual Reality for Creative Resilience, Expression, and Social Integration) as an HCI-oriented conceptual framework for future design and evaluation. The review maps empirical and design-oriented literature on immersive VR, creative engagement, emotional resilience, and social connectedness, distinguishing direct creative-VR evidence from partial clinical, adjacent creative, and contextual sources. The evidence suggests that creative VR may support engagement, perceived agency, emotional expression, and social connectedness, but direct clinical evidence remains limited and preliminary. VR-CREAT translates the mapped evidence into candidate mechanisms, design requirements, testable propositions, and evaluation domains for future prototyping, usability testing, and controlled studies. The framework should therefore be understood as an unvalidated design and evaluation model, not as evidence of clinical effectiveness, cost-effectiveness, or readiness for large-scale implementation.

Computers doi: 10.3390/computers15060383

Authors: Athifah Utami David Mazoteras-Delgado Lucrezia Crescenzi-Lanna

Automatic gesture and movement recognition techniques are mainly used with adults for various purposes in public, clinical, and laboratory settings. Growing interest in this field has led to the increasing application of these methods in child behavioural analysis to serve different societal and educational functions. However, manual human annotation of behaviours remains the predominant method, and only a limited number of studies have explored the use of automatic recognition for children. This review aims to evaluate the rapidly developing techniques of automatic gesture and movement recognition that focus on child behaviour analysis across different settings and for different purposes. More specifically, it analyzes their purposes, target groups, settings, accuracy, and limitations, as well as the ethical issues and data privacy frameworks that should be considered in child-centred AI. Using a systematic review approach following the PRISMA guidelines, this study examines research published between 2021 and 2025 in four databases: Web of Science (WoS), Scopus, PubMed, and IEEE Xplore. From a total of 27 included studies, the findings reveal that automatic gesture and movement recognition is being applied across multiple fields, with consideration of children’s developmental needs. However, a critical gap in technical reporting was identified: fewer than half of the included studies (44%) provided accuracy metrics or clinical validity. Furthermore, evidence of robust ethical safeguards remains limited. To support children’s well-being, future studies must bridge the lab-to-field gap, prioritize natural research settings and enforce ethical and data protection measures.

Computers doi: 10.3390/computers15060382

Authors: Raikhan Amanova Baurzhan Belgibayev Yersaiyn Mailybayev Gulnur Kazbekova Zhadyra Akanova Galiya Mamankyzy Marzhana Amanova Artem Bykov Periuza Pirniyazova Nurzhigit Smailov

This paper proposes a prototype-level hierarchical visual–sensor framework for monitoring the Zamioculcas houseplant in complex indoor environments and supporting adaptive care-mode selection. The proposed framework combines a two-level visual pipeline, consisting of YOLO-based target plant detection and MobileViT-S-based leaf-condition classification, with a Plant Health Index (PHI) and a rule-based decision-support module for integrating visual and IoT-derived indicators. For the detection task, YOLOv8, YOLO12, and YOLO26 were compared, with YOLO26 showing the most balanced performance among the evaluated implementations. To improve robustness in real indoor scenes, negative training samples were added; this reduced the image-level false alarm rate on an independent negative-scene test set from 50.7% to 10.0% and increased specificity from 49.3% to 90.0%. For the second visual level, MobileViT-S achieved an accuracy of 0.9857 and an F1-score of 0.9857 on the independent cropped leaf test subset. To reduce the dependence of this result on a single data split, an additional 5-fold cross-validation experiment was conducted on the full cropped leaf dataset of 847 images, resulting in an accuracy of 0.9858 ± 0.0068 and an F1-score of 0.9853 ± 0.0070. To further address plant-level generalization, an additional unseen-plant validation subset of 60 newly collected cropped leaf images was evaluated, and MobileViT-S achieved an accuracy of 0.9500 and an F1-score of 0.9499. These results support the stability of the leaf-condition classifier within the available data, although larger external validation with strict plant-level and session-level separation remains necessary. In addition, an Arduino-based module-level validation was conducted using a capacitive soil-moisture sensor to verify the proposed sensor-based and Vision–IoT decision rules. The experiment demonstrated that the rule-based layer can distinguish dry, normal, and wet soil states and select conservative care actions depending on both soil moisture and visual-condition input. A brief real-time camera–sensor communication test further confirmed that live camera input, Arduino-based soil-moisture sensing, PHI computation, and care-mode selection can be connected within one decision-support pipeline. The proposed PHI and care-mode selection module are therefore presented as a formalized decision-support layer rather than as a fully validated autonomous irrigation system. Further calibration, actuator integration, and closed-loop validation remain necessary before practical autonomous deployment.

Computers doi: 10.3390/computers15060381

Authors: Mohsen Golafrouz Houshyar Asadi Mohammad Reza Chalak Qazani Anwar Hosen Zoran Najdovski Lei Wei Sam Oladazimi Saeid Nahavandi

Emotion recognition plays a crucial role in human–computer interaction, health monitoring, and affective computing by analysing physiological signals. Despite recent advancements, current research still faces challenges, including the lack of effective fusion strategies for diverse physiological modalities, difficulties in handling high-dimensional feature representations, and limited use of efficient temporal modelling techniques to capture complex emotional patterns. This study proposes a deep learning-based approach that fuses multiple physiological modalities, including Electroencephalography (EEG), Electrooculography (EOG), Electromyography (EMG), Galvanic Skin Response (GSR), Respiratory Rate (RR), Skin Temperature (SKT), and Photoplethysmography (PPG), to improve emotion recognition. Arousal and valence ratings were binarized into two classes (low/high) using a threshold of 4.5, formulating a binary classification problem. In addition to utilising Bidirectional Long Short-Term Memory (Bi-LSTM), the study employs Temporal Convolutional Networks (TCN), a widely used approach for time-series analysis, to efficiently capture temporal dependencies. The proposed model optimises feature selection through channel-wise strategies, incorporates advanced learning rate scheduling, and reduces computational overhead. Furthermore, window-wise, block-wise, and trial-wise evaluation protocols were investigated to assess the impact of temporal information leakage on emotion recognition performance. Using the DEAP dataset for validation, the proposed TCN-based approach achieved classification accuracies of 88.42% for valence and 86.35% for arousal under an overlapping block-wise evaluation protocol, demonstrating improved performance in binary emotion recognition and highlighting the importance of leakage-aware model assessment.

Computers doi: 10.3390/computers15060380

Authors: Frédéric Pattyn Khandakar Rabbi Ahmed Peter Goetz

Feature prioritization in early-stage software startups is a critical yet poorly structured challenge, as prevailing frameworks rely predominantly on expert intuition and fail to exploit patterns latent in historical delivery data and labor market dynamics. This study proposes a deep learning framework that explores labor-market signals as a reproducible proxy for market-driven feature prioritization. The framework encodes two complementary information sources: internal sprint delivery history processed by a Bidirectional Long Short-Term Memory network with attention, and external market signals from LinkedIn job postings processed by a Convolutional Neural Network encoder; the resulting representations are fused via a cross-modal layer to classify job postings as proxies for High or Low feature-priority market demand. The model is evaluated on the publicly accessible LinkedIn Job Postings dataset (2023–2024, approximately 124,000 records) and achieves an Area Under the Receiver Operating Characteristic Curve of 0.961 on the proxy classification task, outperforming classical baselines including gradient Boosted Trees, Random Forest, Support Vector Machine, and Logistic Regression. SHapley Additive exPlanations analysis identifies industry sector and geographic location as the two most influential market-signal predictors. These results suggest that jointly encoding internal delivery dynamics and external market signals offers a promising, scalable decision-support tool to assist startup product teams in data-driven roadmap prioritization, subject to further validation against direct expert priority labels.

Computers doi: 10.3390/computers15060379

Authors: Igor Kalmykov Gennady Slyusarev Vladimir Kopytov Alexander Olenev Tatyana Peleshenko Maria Lapina

Despite significant improvements in vehicle management efficiency achieved through the integration of VANET and Internet of Things technologies, Internet of Vehicles (IoV) networks remain vulnerable to cyberattacks. This is because the wireless data exchange channel in IoV has several vulnerabilities that are exploited to carry out cyberattacks. The article suggested using the symmetric block cipher GOST 34.12-2015 (SBCG) to combat a variety of cyberattacks. This cipher was chosen because it can be efficiently implemented on low-power platforms and offers high cryptographic strength and encryption speed. Furthermore, implementing SBCG in polynomial modular codes (PMCs) enables detection of encryption errors caused by faults in encoder/decoder operation. The scientific novelty of the proposed solution is that it is the first method to increase the fault tolerance of an SBCG encoder, enabling real-time, effective countermeasures against faults caused by both Differential Fault Analysis (DFA) attacks and natural faults. The originality of the solution lies in the integration of cryptographic theory and the theory of constructing correcting modular codes. The goal of this study is to improve the resilience of SBCG encryptors/decoders to faults by using polynomial modular codes. Imparting fault-tolerant properties to SBCG encryption systems implemented in PMC will enable them to effectively mitigate real-time faults arising from both Differential Fault Analysis (DFA) attacks and natural faults.

Computers doi: 10.3390/computers15060378

Authors: Anibal Flores Jose Guzman-Valdivia Saul Huaquipaco Hugo Tito-Chura Ruso Morales-Gonzales Carlos Silva-Delgado Eduardo Flores-Quispe

Global avocado production exceeds 10 million tons annually. Among the diseases affecting avocado fruit, anthracnose is one of the most significant, causing black lesions and fruit decay that can result in yield losses of 20–30%. To facilitate the early detection of anthracnose, this study proposes a computer vision-based approach. A dataset containing 2218 images of Fuerte avocados was first developed, comprising 1730 healthy samples and 488 anthracnose-infected samples after the labeling process. In the experimental phase, several convolutional neural network (CNN) models with varying depths (3, 4, 5, and 6 layers) were designed and evaluated. These models were subsequently integrated into different weighted ensemble configurations, where the best performance was achieved by the ensemble combining all four individual CNNs. The proposed weighted ensemble was compared against widely used state-of-the-art architectures, including VGG-16, ResNet-18, and MobileNetV2. Experimental results demonstrated the effectiveness of the proposed approach, achieving an F1-score of 0.9052, outperforming VGG-16 (0.8283), ResNet-18 (0.7328), and MobileNetV2 (0.7320).

Computers doi: 10.3390/computers15060377

Authors: Awad M. Awadelkarim

The growing sophistication of the internet and web space has spawned highly dynamic, multi-vector cyber threats that cannot be handled by automated detectives and hence the necessity to introduce analyst-oriented, cognitively powerful security analysis apparatus. The character of current visualization-based security frameworks is that they are inclined to deliver data unproactively, fail to engage the dynamic setting, and fail to comprehend the evolving motive of assailants, resulting in subsequent identification and a fractured understanding of coordinated web attacks. The paper introduces a new model of interactive security visualization known as Context-Oriented Visual Exploration of Resilient Threats (COVERT), a hybrid of behavioral context modeling, adaptive visual storytelling, and intent-sensitive interaction. COVERT is dynamically rearranged to the development of threats, patterns of interaction between analysts, and objectives of the possible attacks, which helps in releasing relevant security capabilities gradually. The framework integrates graphical threat flows, attention-directed visual cues, and real-time feedback loops to align system responses to the thinking processes of the analysts. The evaluation of high-scale web traffic and attack simulation dataset indicates that COVERT is much more effective in the multi-stage detection of attacks, false-positive interpretation is minimized, and the investigation period is reduced compared to the visualization infrastructure of the static and semi-interactive infrastructure. According to user studies, there is higher situation awareness, enhanced correlation of distributed events, and enhanced decision-making in complex web intrusion situations, such as advanced persistent threats and web exploitation coordination. Combining contextual intelligence with adaptive interaction and visualization of security, COVERT reveals that intent-based visual analytics may greatly improve internet and web threat detection and analysis systems to support more agile and resilient cyber defense procedures. The proposed COVERT strategy achieved 93% threat-detection rate, the false positives were reduced to 6%, the response time of the analysts was reduced to 140 s, and the situational awareness was increased to 88%.

Computers doi: 10.3390/computers15060376

Authors: Oluwaseyi Oladejo Ahmed Abdelmoamen Ahmed

The exponential growth of cyber threats in modern digital infrastructure demands advanced detection systems that adapt to evolving attack patterns. Traditional cybersecurity approaches struggle with dynamic threats, requiring extensive labeled datasets and retraining for each new category. This paper presents a comprehensive transfer learning framework for cybersecurity threat detection, leveraging the CICIoMT dataset as a benchmark to enhance detection capabilities across heterogeneous cybersecurity environments. We propose a machine learning (ML)-enabled framework that employs systematic feature alignment, hybrid class balancing, and multi-algorithm evaluation using machine learning models, including Multi-Layer Perceptron (MLP), Support Vector Machine (SVM), Random Forest (RF), Gradient Boosting, and XGBoost. The proposed approach addresses the critical challenges of data scarcity and domain heterogeneity in cybersecurity by enhancing feature engineering with cybersecurity-specific features, statistical aggregations, and PCA embeddings. Extensive experimental evaluation across two target datasets (CICIoT and IoT-23) demonstrates both the exceptional successes and critical limitations of cross-domain transfer learning in cybersecurity. The framework achieved outstanding performance on domain-compatible datasets, with RF reaching 99.0% accuracy on CICIoT, Gradient Boosting achieving 98.9%, and XGBoost delivering 98.4%, demonstrating exceptional knowledge transfer from medical IoT to smart home IoT environments. However, transfer learning to IoT-23 was unsuccessful (50% accuracy, equivalent to random guessing), revealing that feature domain difference, where identical attack labels encode fundamentally different behavioral patterns, prevents effective knowledge transfer despite nominal class overlap. This research makes significant advances in adaptive cybersecurity systems by providing a rigorous evaluation of both the successes and limitations of transfer learning. This work demonstrates that ensemble methods (RF, XGBoost, and Gradient Boosting) achieve superior cross-domain performance compared with neural networks on compatible domains, while also revealing fundamental challenges when the source and target domains differ in their feature spaces.

Computers doi: 10.3390/computers15060375

Authors: Lu Xu

Inventory systems are typically evaluated using aggregate performance metrics such as out-of-stock and average inventory. In supply chain management, it is important to understand the underlying reasons for a period’s performance—specifically, how previous inventory management decisions, such as order placement, lead to the result and what their contributions are. Traditional methods are often restrictive and cannot be applied to broader cases. This paper proposes a Shapley-based decomposition framework that attributes the realized performance gap between the observed inventory policy and optimized reference policy to individual decisions. A numerical experiment on a simulated finite-horizon periodic-review inventory system with stochastic demand and lead time is conducted to illustrate the basic idea of the method. Compared to traditional methods, the proposed approach directly explains a realized benchmark-relative performance difference and is applicable to integer-constrained, non-differentiable, and simulation-based inventory systems. It enables transparent inventory management performance evaluation and effective root-cause analysis.

Computers doi: 10.3390/computers15060374

Authors: Milena Nikolić Miloš Stojanović Marina Marjanović

Online hotel reviews provide a major source of information for understanding guest satisfaction, yet rating prediction remains difficult because review text is often short, highly skewed toward positive scores, and affected by inconsistent or repetitive content. This study presents a framework designed to prevent information leakage in hotel rating prediction. The empirical analysis uses 26,675 Booking.com reviews covering 819 hotels. After minimal cleaning, 26,384 reviews are retained and split chronologically before any vectorizer fitting, reliability scoring, or model training. The protocol ensures that the test set is never filtered and that no rating-derived reliability signal is used to construct training features. The study evaluates TF-IDF Ridge baselines, recurrent neural models, and transformer architectures including DistilBERT, BERT, RoBERTa, MobileBERT, and TinyBERT. Evaluation is performed on the same unfiltered chronological test set using MAE, RMSE, R2, rounded rating accuracy, accuracy within one rating point, low score recall, rating group errors, and confusion matrices. DistilBERT achieves the strongest overall performance, with MAE = 0.4370, RMSE = 0.6979, and R2=0.8217, while BiLSTM models show stronger sensitivity to low rating reviews. Additional analyses include language composition auditing, unseen hotel generalization, reliability threshold sensitivity, and forecasting of future hotel rating dynamics across 7-, 30-, 60-, and 90-day horizons. The results show that reliable hotel rating prediction requires both expressive language models and careful evaluation protocols that separate review reliability analysis from filtering based on the target rating.

Computers doi: 10.3390/computers15060373

Authors: Aigul Tungatarova Gaukhar Borankulova Aslanbek Murzakhmetov Bakhyt Yeraliyeva Saltanat Dulatbayeva Samat Bekbolatov Balzhan Turarova

Efficient water resource management is critically important for arid regions such as southern Kazakhstan. This paper presents a hybrid Internet of Things (IoT) and Vision-based Internet of Things (VIoT) architecture for real-time monitoring of water levels in irrigation channels. The proposed system integrates an ultrasonic water-level sensor, an IP camera with edge-based computer vision processing on a Raspberry Pi, wireless communication, an autonomous solar power supply, and discharge estimation using Manning’s equation. The VIoT subsystem applies image processing techniques, including gauge calibration, Canny edge detection, and pixel-to-metric conversion, to automatically estimate water level from captured video frames. Water-level measurements obtained from IoT sensors and video-based analysis are combined through synchronised data fusion to improve monitoring accuracy and reliability. The hybrid approach leverages the complementary strengths of IoT and VIoT by combining continuous quantitative sensing with visual verification capabilities. Field experiments conducted on the Merke River in the Zhambyl region of Kazakhstan over a 14-day observation period demonstrated stable real-time operation with RMSE = 0.311 cm, MAE = 0.279 cm, and Pearson r = 0.99 between the ultrasonic sensor and the vision-based estimates. Sensitivity analysis indicated that water level is the most influential parameter in Manning-based discharge estimation, confirming the importance of accurate level detection. The proposed system improves reliability by cross-checking independent data sources, making it applicable to monitoring water levels in agricultural regions.

Computers doi: 10.3390/computers15060372

Authors: Peng Dai Xiaoqin Ma Yanyang Zhao Yunzhan Gong

Continuous Integration pipelines rely on large-scale automated testing to support rapid releases. However, flaky tests exhibit non-deterministic outcomes under an identical code and configuration, substantially increasing rerun costs and hindering fault localization. Existing approaches struggle to uniformly model heterogeneous runtime evidence and its multi-relational structure in CI environments, which limits cross-project generalization and interpretability. To address this gap, this paper presents HgtFlaky, a runtime-evidence-centered multi-view heterogeneous graph learning framework. A Unified Event Model is introduced to normalize heterogeneous CI artifacts into semantically consistent event quadruples, and a heterogeneous execution graph is then constructed to capture testing entities and multiple relation types. Based on the HEG, three complementary views are derived to characterize run-level, test-level, and thread-level flaky behaviors. A heterogeneous graph Transformer is further adopted to jointly encode the multi-view graph instances and learn transferable test-level representations for flaky/non-flaky prediction. Experiments on two benchmark datasets, FlakeFlagger and IDoFT, show that HgtFlaky achieves strong and stable performance. Under 10-fold cross-validation, it obtains an F1-score of 83% on FlakeFlagger and 98% on IDoFT. Under per-project validation on FlakeFlagger, HgtFlaky achieves 78% Precision, 89% Recall, and 81% F1-score, outperforming Flakify by 8 percentage points and FlakeFlagger by 74 percentage points in F1-score.

Computers doi: 10.3390/computers15060371

Authors: Mohammad Rasras Iuliana Marin

One of the remaining challenges in deploying 3D CNN models in resource-constrained environments is the high computational demand. In this paper, we design three lightweight architectures that have distinct spatiotemporal topologies, namely, Lite-R21D, Lite-MC3, and Lite-LF, to reduce computational cost. However, these compact models have restricted representational capacity, which consequently limits their ability to capture complex spatiotemporal features. To overcome this, we employ Knowledge Distillation (KD) and further investigate hybrid combinations of response-based, spatiotemporal attention, and intermediate feature alignment paradigms. By analyzing knowledge transfer across these diverse architectures, our experiments on UCF101 and HMDB51 demonstrate that combining these distillation configurations consistently outperforms single KD methods, resulting in a substantial increase in accuracy across all Student models. Our optimal hybrid setup achieves 92.07% accuracy on UCF101 and 65.56% on HMDB51, compared to the Teacher’s 94.74% and 69.48%, reducing the accuracy gap to only 2.67% and 3.92%. These gains are achieved alongside significant efficiency improvements. The proposed models operate with up to 87% fewer parameters and an 89% reduction in Floating-Point Operations (FLOPs), achieving 6.7× faster inference. Our findings highlight that hybrid distillation is an effective approach for transferring and utilizing complex spatiotemporal knowledge in lightweight models.

Computers doi: 10.3390/computers15060370

Authors: Symon Fumu Nyalugwe Okuthe P. Kogeda Robert Hans

Traffic congestion poses a significant challenge in urban environments. The use of digital techniques has emerged as a pivotal trend, as it offers substantial safety to and mitigates stress and frustration for road users. The purpose of this survey was to explore the current approaches and digital techniques for managing traffic congestion. We address this through a systematic literature review (SLR) approach by adopting PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. We began by exploring the key techniques of topological data analysis (TDA), machine learning (ML) and deep learning (DL) for modeling urban traffic prediction. We evaluated the robustness of the topological data analysis technique (Persistent Homology (PH)) against deep learning frameworks (Graph Convolutional Neural Networks (GCNNs)). We found that each framework has its own strengths and weaknesses, and neither of the frameworks independently provides a complete solution. PH may offer richer structural insights and robustness to noise but may struggle with direct predictive implementation, while deep learning models do better at extracting dynamic predictive patterns but are assumed to lack interpretability and generalizability. Therefore, the integration of multiple techniques, either PH with stacking ensemble methods or deep learning with stacking ensemble methods, can improve prediction and generalization of the model while at the same time reducing over-reliance on local graph assumptions. Future research should focus not only on performance metrics or methods but also on explainability, transferability, adaptability across heterogeneous road environments and computational cost.

Computers doi: 10.3390/computers15060369

Authors: Nureni Ayofe Azeez Abdullateef Akorede Ademoye Oluwatobi Sunday Malomo Omotolani Okerinde Mary Damilola Seun Aaron Charles VanDer Vyver

This study investigates data augmentation as a strategy for addressing dataset scarcity in Internet of Medical Things (IoMT) cybersecurity and improving intrusion-detection system performance. Four augmentation methods—Rule-Based, Tabular Variational Autoencoder (TVAE), Conditional Tabular Generative Adversarial Network (CTGAN), and Gaussian Copula—were applied to two publicly available IoMT datasets (ECU-IoHT and WUSTL-EHMS) to generate augmented training data with differing class distributions and feature characteristics. Eleven machine learning algorithms were evaluated using Matthews Correlation Coefficient (MCC), F1-score, accuracy, and error-based metrics. Results showed consistent performance improvements across all evaluated models relative to the baseline datasets. The Rule-Based method produced the strongest overall results, achieving the highest MCC (0.9757), F1-score (99.19%), and accuracy (99.18%) with LightGBM, alongside low false-positive and false-negative rates. Among the generative approaches, TVAE delivered the strongest overall practical performance (F1-score = 96.94%, accuracy = 96.92%), while CTGAN achieved a marginally higher MCC (0.9047) and also produced competitive results with balanced class representation. Gaussian Copula generated the weakest overall outcomes, primarily due to highly skewed class distributions. Traditional models, such as Logistic Regression and Naive Bayes, recorded the largest relative gains, indicating that augmentation can substantially improve simpler classifiers in data-scarce environments. Overall, the findings demonstrate that augmentation quality depends not only on dataset expansion, but also on preserving class balance, feature diversity, and realistic traffic relationships. These results provide practical guidance for strengthening IoMT intrusion-detection systems in healthcare environments.

Computers doi: 10.3390/computers15060368

Authors: Sahar Saadallah Ahmed Mohand Lokman Al dabag

Structured Query Language injection (SQLi) remains a persistent threat to web applications due to the obfuscation, diversity, and evolving structure of malicious payloads, which limit the effectiveness of conventional rule and signature-based Web Application Firewalls (WAFs). Although prior studies have reported high detection performance using individual feature extraction methods or offline classification models, limited work has addressed deployment-oriented SQLi prevention through an integrated real-time inspection framework. This paper proposes a Machine Learning (ML)-based SQLi detection and prevention framework that combines hybrid feature representation, supervised dimensionality reduction, Genetic Algorithm (GA)-based hyperparameter optimization, and real-time WAF validation. Multiple public SQLi datasets were merged, cleaned, and deduplicated to improve exposure to diverse query patterns. SQL queries were encoded using Term Frequency–Inverse Document Frequency (TF-IDF), Word2Vec, and FastText features, which were fused and transformed through a Supervised Autoencoder into a compact discriminative representation. GA was then employed to optimize multiple classifiers, including Random Forest (RF), Support Vector Machine (SVM), Decision Tree (DT), and Multi-Layer Perceptron (MLP). The MLP achieved the best overall performance, with an accuracy of 0.998681. The optimized model was deployed within a lightweight Flask-based WAF for real-time Hypertext Transfer Protocol (HTTP) request inspection and malicious input blocking. SQLMap v1.8.4-based robustness testing and runtime analysis demonstrate that the proposed framework provides effective SQLi prevention with practical deployment efficiency beyond conventional offline benchmark evaluation.

Computers doi: 10.3390/computers15060367

Authors: Alexandru Răzvan Căciulescu Matei Bădănoiu Răzvan Rughiniș Dinu Țurcanu

Java middleware platforms expose powerful management functions through HTTP-accessible interfaces such as Jolokia. This article discusses the analysis of CVE-2023-50780 in Apache ActiveMQ Artemis by framing the vulnerability as a management-plane state-transition problem rather than as a set of isolated exploit recipes. We analyze three remote-code-execution paths that combine Jolokia-accessible MBeans with Log4J2 configuration mutability, Artemis filesystem and deployment semantics, broker or web-server restart behavior, and, in one vector, the Java DiagnosticCommand interface. The study defines a formal attacker model; separates demonstrated preconditions from deployment-dependent assumptions; compares the three vectors across required privileges, network dependencies, writable artifacts, execution triggers, reliability, detection opportunities, and mitigations; and evaluates defensive controls at the level of the exploit stage they interrupt. The paper also clarifies the responsible-disclosure context and reduces operational payload detail in favor of defender-oriented evidence, validation tables, and architectural analysis. The resulting contribution is a reproducible but bounded case study of how legitimate administrative operations can compose into code execution when management interfaces are exposed without sufficient privilege separation, MBean restriction, filesystem hardening, and upgrade controls.

Computers doi: 10.3390/computers15060366

Authors: Kassim B. Diallo Moulay A. Akhloufi

The field of pose estimation is a major problem in computer vision, enabling the direct transformation of an input image into a hierarchical representation of the human skeleton for application in the fields of virtual/augmented reality and human–machine interaction tasks. Research in this field has exploded between 2018 and 2025, with traditional taxonomies such as 2D versus 3D or top-down versus bottom-up no longer sufficient to capture the essence of the evolution of ideas. To solve this problem, we propose a conceptual review in the field of pose estimation, focusing on the intellectual evolution of methods and architecture rather than the standard flat classifications of papers. We divide recent advances into five structural pillars: Representation, which traces the evolution from pixel coordinate regression to heatmaps and probabilistic representation; Architecture, which analyzes the transition from multi-stage CNNs to transformers and state space models (SSMs); Ambiguity and Generalization, which analyzes how self-supervised, uncertainty-aware, and diffusion models address 3D depth ambiguity, occlusion, and domain gaps by modeling multiple plausible poses and reducing dependence on fully supervised in-the-wild 3D labels; Context Extension, which covers temporal dynamics, multi-view fusion, and potential sensors; and Applications, which links algorithms to efficiency, privacy, and foundation models. By providing an in-depth detailing of these pillars, we provide a unified view of the evolution of research paradigms that define human pose estimation and enable the identification of future problems and solutions in pose estimation and human-centered tasks.

Computers doi: 10.3390/computers15060365

Authors: Indira Karymsakova Dinara Kozhakhmetova Dariga Bekenova Lili Nurliyana Abdullah Aleksandr Zolotov Dinara Shyrynkhanova Lazzat Kydyralina Alina Bugubayeva

Lung cancer is still one of the major contributors to cancer-related mortality. Treatment effectiveness depends directly on early diagnosis and accurate prediction of disease progression. This study focuses on the application of classical machine learning algorithms —including Logistic Regression, Support Vector Machine, Random Forest, and XGBoost —for questionnaire-based lung cancer risk assessment. A comparative study of six machine learning algorithms used to predict lung cancer based on survey data was conducted. According to the ROC-AUC metric, Random Forest model achieved the best result. In terms of F1-score and precision for the positive class, the support vector machine (SVM) demonstrated the highest efficiency. However, gradient boosting provided the most balanced results across key clinical metrics. SHAP (SHapley Additive exPlanations) analysis identified three preliminary predictors potentially associated with lung cancer risk in this sample: seeing a pulmonologist, unexplained weight loss, and loss of appetite. These preliminary results are consistent with clinical observations and suggest the potential interpretability of ensemble machine learning approaches in medical diagnostics, though confirmation on larger datasets is required. Overall, the preliminary results suggest the potential of using machine learning methods for lung cancer screening based on questionnaire data, particularly with an expanded training sample. Potential applications of early lung cancer diagnosis are discussed. Further research in this area will be conducted in conjunction with image recognition of computed tomography scans.

Computers doi: 10.3390/computers15060364

Authors: Said Zulfigarzada Aysun Gadirli Javid Karimov Danas Cerneckas Roma Rackiene Mindaugas Azubalis

Industrial automation systems are increasingly cyber-physical, interconnected, and software-dependent, which expands both their operational capability and their cybersecurity exposure. This article reports a systematic literature review, conducted following the PRISMA 2020 guidelines, of cybersecurity requirements and certification standards in industrial automation, with emphasis on Industrial Control Systems (ICS), Supervisory Control and Data Acquisition (SCADA), Programmable Logic Controllers (PLCs), and Industry 4.0 contexts. From 3570 records identified across five academic databases, 75 studies were retained after duplicate removal, title and abstract screening, and full-text eligibility assessment. The included studies were analyzed along three dimensions: cybersecurity requirements, standards and certification, and application context. Quantitative synthesis shows that network segmentation, intrusion detection, secure communication, access control, lifecycle security, and safety–security coordination are the six most frequently emphasized requirement categories, and that ISA/IEC 62443, ISO/IEC 27001, NIST SP 800-82, and NERC-CIP are the four dominant certification frameworks. The review identifies four critical gaps between technical cybersecurity requirements and certification practice and proposes an integrated mapping framework linking requirement categories, standards, and application contexts. The findings indicate that effective industrial cybersecurity assurance depends on a layered compliance architecture rather than on dependence on any single framework.

Computers doi: 10.3390/computers15060363

Authors: Shuaiming Chen Kerui Xie Mengting Yuan

Ray tracing is a fundamental technique in computer graphics that requires highly efficient algorithms to achieve real-time rendering performance. A key bottleneck in this process is the traversal of Bounding Volume Hierarchies (BVHs). This paper introduces BMA (BMH Mapping Algorithm), where BMH stands for a Bounding-Volume Mapping Heuristic. BMA is a sampling-based algorithm designed to accelerate BVH traversal by leveraging probabilistic relationships between rays and BVH nodes. Through a convolution-like operation, the algorithm estimates traversal probabilities, enabling rays to bypass redundant or low-probability nodes. Experimental results demonstrate a significant acceleration in BVH traversal performance, with up to a 150% improvement on test scenes from the McGuire Computer Graphics Archive. The method maintains high visual fidelity, achieving PSNR values above 35 and SSIM values above 0.9. BMA is also compatible with existing acceleration and optimization techniques, making it a practical addition to real-time ray tracing pipelines.

Computers doi: 10.3390/computers15060362

Authors: Narjes Lassoued Imen Filali Ridha Ejbali

The evolution of 5G networks has introduced new challenges in securing mobile infrastructures against increasingly sophisticated cyber threats. Intrusion detection in such environments has been widely studied using traditional datasets such as the Canadian Institute for Cybersecurity Intrusion Detection Systems CICIDS2017, the University of New South Wales-Network Behavior UNSW-NB15, and The Network Security Laboratory-Knowledge Discovery in Databases NSL-KDD; however, these benchmarks lack the architectural complexity and protocol diversity inherent to 5G networks. More recent research has adopted the 5G-NIDD dataset (5G Network Intrusion Detection Dataset), which provides realistic traffic generated from a live 5G testbed, including various attack scenarios targeting MEC servers and core network components. Nevertheless, existing works using 5G-NIDD often focus on limited subsets of attacks, rely on unsupervised or federated learning approaches, and lack comprehensive evaluations of supervised learning models. In contrast, this study leverages the entire 5G-NIDD dataset, encompassing all available attack scenarios, and conducts a systematic comparison of multiple supervised learning algorithms. A systematic evaluation of supervised learning algorithms is conducted using key performance metrics such as accuracy, precision, recall and F1-score to identify the most effective model for intrusion detection in 5G environments. Specifically, this study focuses on four supervised learning algorithms, K-Nearest Neighbors (KNNs), Support Vector Machines (SVMs), Logistic Regression (LR), and Naive Bayes (NB), to determine not only which achieves the highest detection accuracy but also which offers the best balance between predictive performance and computational efficiency in realistic 5G environments. To assess robustness and adaptability, the proposed models are further validated on two widely used benchmark datasets, namely CICIDS2017 and UNSW-NB15, as part of an extended analysis. This cross-dataset evaluation highlights each algorithm’s strengths and limitations under diverse network traffic conditions and attack scenarios. The results aim to validate the applicability of supervised learning approaches to intrusion detection in next-generation network infrastructures, while also emphasizing the importance of balancing predictive accuracy with computational efficiency for real-world deployment.

Computers doi: 10.3390/computers15060361

Authors: Saeed Mohsen Saada Khadragy Norah Alnaim Noorah Albehaijan Ahmed F. Ibrahim

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder that requires early and accurate diagnosis to improve patient outcomes. In this paper, an attention-enhanced hybrid deep learning (DL) framework is proposed that combines Convolutional Neural Network (CNN) and Swin Transformer (Swin-T) architectures for multi-class Alzheimer’s classification. The proposed model integrates an attention mechanism to enhance feature representation and improve classification performance. Experiments are conducted on a dataset containing three classes: Mild Demented, Very Mild Demented, and Non-Demented. To improve the model’s generalization, data augmentation techniques are applied to enhance the model’s performance. Additionally, three explainable artificial intelligence (XAI) techniques are employed, including Grad-CAM++, Integrated Gradients, and Saliency maps, to interpret the model’s predictions and to provide visual insights into decision-making processes. The proposed attention-enhanced hybrid CNN–Swin-T model achieves a testing accuracy of 99.92% and reaches 99.71%, 99.73%, and 99.72%, for precision, recall, and F1-score, respectively. The hybrid CNN–Swin-T with attention outperforms three implemented models: baseline CNN, standalone Swin-T, and hybrid CNN–Swin-T. The explainability results validate the proposed model’s focus on relevant regions, increasing trust in automated diagnosis systems. Finally, a comparative analysis with an ablation study is presented to demonstrate that the integration of the attention mechanism with a hybrid CNN–Swin-T architecture leads to the highest performance and more reliable predictions compared to the other three models.

Computers doi: 10.3390/computers15060360

Authors: Uğurcan Sert Esra Ersoy Ömür Tosun Irmak Hatıpoğlu

Vaccination stands as a preeminent public health measure in the fight against infectious diseases, with a proven track record of significantly reducing morbidity and mortality rates. However, the presence of vaccine hesitancy and misinformation, particularly evident during the course of the pandemic, has emerged as a significant challenge. The present study analyzes public perceptions of vaccination by examining YouTube comments on 215 vaccine-related videos, which total over 94,000 comments. Employing advanced topic modeling techniques, such as Hierarchical Dirichlet Process (hLDA), Latent Semantic Analysis (LSA), and Non-Negative Matrix Factorization (NMF), the study identifies key themes, including vaccine safety, side effects, pharmaceutical ethics, and public trust in healthcare authorities. The findings indicate that debates frequently center on political, social, and scientific concepts. Vaccine hesitancy has emerged as a pervasive global phenomenon that transcends cultural boundaries. The dissemination of misinformation regarding the efficacy of vaccines and the safety of treatments, such as ivermectin, is a prevalent phenomenon on social media platforms. This poses significant challenges to public health efforts. The subjects of child vaccination and parental standpoints are also recurring topics of concern. This study underscores the pivotal function of digital platforms such as YouTube in influencing public attitudes regarding vaccination. This underscores the necessity for targeted communication strategies, advanced digital literacy, and proactive policies by social media platforms to address misinformation and promote evidence-based information. Such precautions are imperative to sustaining elevated vaccination rates and safeguarding public health in the digital age.

Computers doi: 10.3390/computers15060359

Authors: Rajagopal Kumaraswamy V. Sheeja Kumari N. Muthuvairavan Pillai R. H. Aswathy Vijayalakshmi Ramakumar Indra Neel Pulidindi

Precise segmentation of kidney tumors in medical images is crucial for diagnosis, treatment planning, and prognosis assessment. In this work, we present a newly proposed hybrid deep learning model that combines the merits of U-Net and the Swin Transformer architectures in order to enhance the segmentation performance. Although U-Net has great spatial localization ability thanks to the encoder–decoder structure, which works in a hierarchical way, it is still difficult to capture global context well. The Swin Transformer instead captures long-range dependencies and assists in local detail extraction, while attention pooling might also smear fine boundary details. This motivates our hybrid integration. To attempt to resolve these issues, we extend U-Net with the Swin Transformer blocks in the backbone encoder path in order to efficiently perform multi-scale semantic feature extraction while preserving structural consistency. We trained and cross-validated the model on the publicly available Kidney Tumor Segmentation Challenge 2021 (KiTS21) dataset with extensive data augmentation as well as custom loss functions to address class imbalance and boundary obscureness. Experiments demonstrated that it achieved better performance when compared with the solo models, seeking a similar multi-task learning objective on not only U-Net and the Swin Transformer but also other baseline architectures in terms of the average Dice similarity coefficient (average DSC), intersection over union score (IoU) and Hausdorff distance. The proposed model achieved a Dice similarity coefficient (DSC) of 0.91, an IoU of 0.87, a PR-AUC of 0.89, and an overall voxel-wise accuracy of 98%, demonstrating robust and precise kidney tumor segmentation across varying tumor sizes and shapes. Moreover, the integrated solution is more robust and generalizes better, particularly in challenging cases with diverse anatomical variations. These findings demonstrate the power of Transformer-based hybrid models for medical image segmentation. Our results have positive implications for the design of computer-aided diagnostic systems and their association with other prevalent medical imaging tasks besides organ-specific or pathology-focused tasks.

Computers doi: 10.3390/computers15060358

Authors: Mohammad Alhusban

Building Information Modeling (BIM) implementation is increasingly adopted in the architecture, engineering, and construction (AEC) industries. However, its integration into the academic curricula in developing countries remains limited. Therefore, this study aims to investigate the barriers to integrating BIM into the curricula of civil engineering in Jordanian higher education institutions (HEIs). A quantitative approach was used, including Exploratory Factor Analysis (EFA) and Partial Least Squares Structural Equation Modeling (PLS-SEM). The data was collected from 102 respondents, including industry professionals and academics. Six key barrier constructs were identified: support, standards, delivery, resources, knowledge, and infrastructure and security. Altogether, they explain 66.896% of the BIM integration barriers. The results of the structural model indicate that institutional and governmental support is the most critical barrier (β = 0.486), followed by the lack of standards (β = 0.206) and curriculum-delivery constraints (β = 0.166). Other barriers, including infrastructure and security-related factors, knowledge gaps, and resource limitations, were found to have statistically significant effects on BIM integration. The findings revealed that the barriers to integrating BIM into civil engineering curricula in Jordanian HEIs are institutional and systemic rather than purely technical or resource-based. This study contributes to the BIM education literature by developing one of the first empirically validated PLS-SEM models to investigate barriers to integrating BIM curriculum in Jordan and in developing countries. This research is distinct from previous descriptive studies by prioritizing the institutional, technical, and curricular barriers to the integration of BIM into civil engineering education. Practically, the research provides a specific roadmap for Jordan to integrate BIM into curricula through improving the collaboration between HEIs and the Jordan Engineering Association, strengthening the accreditation standards, enhancing the support of the government for digital construction education, and endorsing the partnerships between HEIs and the industry to align the graduates with the needs for digital transformation of the construction sector in Jordan.