Computers doi: 10.3390/computers15090569

Authors: Marian Emmanuel Okon David Austria Javonte Williams Tia Smith Aiyana Jones Micheal Olaolu Arowolo

State-level health resource systems need precise and timely prediction models, but they are plagued by the ongoing problem of deteriorating model performance because of constantly shifting data distributions (data drift). Predicting uncommon but important events like sickle cell crisis is a classification task where this problem is most noticeable. This paper presents the Evolving AI-Driven Ensemble Learning Framework, which blends novelty detection using the F1-score with sophisticated ensemble approaches (stacking XGBoost, Deep Neural Network, and Random Forest with a meta-learner). Complex, high-dimensional health data are handled using sophisticated feature engineering techniques, such as automated feature selection via evolutionary algorithms and meta-learning (MAML). We empirically assessed a reactive retraining technique that was improved by ensemble stacking and simulated real-time data drift. After five retraining cycles, the improved ensemble and feature engineering showed significant performance improvements over the initial model, achieving substantial improvements: F1-score improved from 0.1250 to 0.9734 (an absolute increase of 0.8484, representing a 678.7% relative improvement), recall from 0.0714 to 0.9767 (an absolute increase of 0.9053), and precision from 0.5000 to 0.9702 (an absolute increase of 0.4702). The framework maintained high specificity (0.9700) and demonstrated outstanding discriminative performance with an AUC-ROC of 0.9909 (an 8.5% improvement). The model’s strong predictive capacity was confirmed by improvements in the Matthews Correlation Coefficient from 0.1000 to 0.9467 (846.7% improvement) and Cohen’s Kappa from 0.0800 to 0.9467 (1083.3% improvement). Model transparency in pipeline development is now made possible by a fixed runtime issued in the SHAP explainability layer. The efficiency of the framework is empirically validated by this study, showing that automated feature engineering and optimized ensemble learning greatly increase model stability and preserve remarkable accuracy for minority classes in complicated data contexts.

Computers doi: 10.3390/computers15090568

Authors: Sokol Shurdhi Eglantina Zyka Luan Bekteshi

Commercial operations across trading, import/export, logistics, and technology distribution are being reshaped by the convergence of artificial intelligence (AI), machine learning, multi-agent robotics, and Extended Reality (XR). Small and Medium-sized Enterprises (SMEs) feel this shift acutely: they face the same pressures as their larger competitors; labor shortages, high-SKU inventories that resist tidy categorization, narrow margins, and customer expectations set by Amazon-grade fulfilment, but rarely command the capital or the structured warehouse environments that make industrial automation straightforward. Existing frameworks for AI-driven automation and digital twins have been developed primarily for large-scale industrial settings and do not account for the capital, infrastructure, and organizational constraints specific to SMEs, leaving a gap in SME-scoped integration models. This research addresses that gap by asking how AI-driven robotics and immersive technologies can be integrated to optimize commercial workflows in SMEs operating in dynamic logistics and trading environments. The proposed framework is grounded in Sociotechnical Systems Theory, which treats technology and organizational workflows as jointly designed and mutually adapting, and follows a Design Science orientation in which the architecture itself is constructed as an evaluable artifact rather than a purely descriptive model. Methodologically, the study conducts a narrative synthesis of literature on embodied AI, computer vision, digital twins, VR training, and AR-assisted operations, combined with workflow analysis to identify where SMEs lose the most time and money. These are translated into a four-layer system architecture (perception, cognition, execution, integration) deployed through a four-phase implementation model: needs assessment, digital-twin and VR pre-training, controlled hardware pilot, and AR-supported scaling. The contribution of the study is twofold: conceptually, it brings together several technologies that are often discussed separately in the literature, while focusing specifically on the needs and constraints of SMEs while practically, it proposes a phased roadmap that can help SMEs adopt these technologies gradually, reducing both financial and operational risks. The approach also emphasizes human–robot collaboration rather than replacing human workers. The study does not include experimental validation, it presents a conceptual architecture and implementation roadmap consistent with a Design Science artifact-construction stage that can serve as a basis for empirical testing in real commercial environments.

Computers doi: 10.3390/computers15090566

Authors: Eduardo Martínez-Duque Genaro Daza-Santacoloma David Cárdenas-Peña

Hypertension is a major risk factor for cardiovascular disease, yet cuffless blood pressure monitoring remains challenging because most existing methods rely on intermittent cuff-based measurements or multimodal physiological signals. This work proposes PPG-FusionNet, a dual-branch deep learning architecture for simultaneous systolic and diastolic blood pressure estimation using only photoplethysmography (PPG) signals. The model combines a local dilated convolutional encoder and a global AutoCorrelation encoder operating on a shared patch embedding, whose representations are integrated through cross-attention fusion and optimized with a constrained dual-head regression objective. The model was trained and evaluated on the PulseDB benchmark using Bayesian hyperparameter optimization and systematic ablation studies to assess each architectural component. PPG-FusionNet achieved mean absolute errors of 7.46 mmHg for systolic blood pressure and 4.72 mmHg for diastolic blood pressure, with near-zero mean errors and compliance with the ANSI/AAMI standard and BHS Grade B for diastolic estimation. Ablation experiments revealed that trend-seasonal decomposition, despite its success in long-horizon forecasting, degraded performance on short PPG windows, whereas cross-attention fusion and AutoCorrelation improved estimation accuracy. These results demonstrate that heterogeneous dual-branch representation learning provides an effective, scalable framework for cuffless blood pressure estimation from a single PPG sensor.

Computers doi: 10.3390/computers15090567

Authors: Xiaoya Zuo Zhanxu Cui Rugui Yao Ye Fan Margulan Ibraimov

Automatic Modulation Recognition (AMR) is a key technology for spectrum sensing and signal demodulation in cognitive radio systems. Most existing AMR methods use raw In-phase/Quadrature (I/Q) data or direct transformations of I/Q data as neural network inputs, while the nonlinear interaction between signal amplitude and phase is not explicitly modelled. This limitation can constrain recognition performance in complex channels and high-order modulation schemes. To address this issue, an AMR method based on Amplitude/Phase (A/P) interaction enhancement is proposed. An Amplitude–Phase Interaction Autoencoder (APIAE) with a residual transition module is designed to learn amplitude–phase interaction features in the polar coordinate domain through unsupervised reconstruction, and the learned interaction features are concatenated with the raw A/P data to form an enhanced three-channel representation for downstream classifiers. Experiments on a subset of RadioML2018.01A show that all five evaluated backbone networks benefit from the proposed representation. Across four independent runs, GRU-Enhanced attains a mean best accuracy of 98.72±0.20%, CLDNN achieves a 7.39% relative gain in overall mean accuracy, and the recognition rates of 16QAM and 64QAM remain markedly improved at 4 dB. The classifier-side input adaptation increases the parameter count by less than 0.5% for all evaluated backbones, while the standalone APIAE front-end contains approximately 7.35 million parameters. The additional end-to-end single-sample inference latency relative to the raw backbone does not exceed 12 μs in the tested environment.

Computers doi: 10.3390/computers15090565

Authors: Dimitra Chasanidou Michail Kalogiannakis Natassa Raikou Georgina Stavropoulou Eleftheria Beazidou

Artificial Intelligence (AI) literacy has emerged as a critical 21st-century competence, yet its integration within interdisciplinary STEAM (Science, Technology, Engineering, Arts, and Mathematics) education remains fragmented. This study addresses this gap by investigating how AI literacy is conceptualized, implemented, and evaluated within STEAM learning environments. Following the PRISMA guidelines, a systematic literature review was conducted on empirical studies published between 2015 and 2025 to examine pedagogical approaches, methodological designs, AI technologies, and arts disciplines. An analysis of 32 studies revealed key insights into the development of AI literacy within STEAM education: (1) AI literacy is most often fostered through active and interdisciplinary learning rather than through lecture-based instruction alone, (2) AI serves both as a subject of study and as a tool for learners to create, design, investigate, and solve problems, (3) arts integration in AI-STEAM education typically support broader learning objectives and less frequently is assessed as distinct learning outcome, (4) ethical and societal aspects of AI literacy receive less attention than technical and computational skills, (5) the predominance of non-formal settings, short interventions, and context-specific studies indicates limited evidence on long-term progression, scalability, and sustained outcomes. The review contributes to the emerging field of AI-STEAM education by mapping empirical research from the past decade and developing two analytical tools: a five-cluster framework for learning objectives and a four-cluster framework for arts integration in AI-STEAM education. It concludes with a multi-level synthesis that identifies key educational and technological implications for the design, implementation, and future development of the field.

Computers doi: 10.3390/computers15090564

Authors: Iván Miguel García-López Nicia Guillén-Yparrea

Open language models are increasingly considered for institutional decision-support tasks in higher education, including policy interpretation, academic advising, administrative summarization, and quality-assurance workflows. However, their reliable deployment requires more than model availability: it depends on cloud-native orchestration, retrieval quality, evidence grounding, refusal behavior, monitoring, and governance controls. Following a design-science research approach, this paper presents an architectural artifact for deploying open language models in higher education decision support. The artifact operationalizes institutional reliability as a multidimensional construct composed of contextual accuracy, answer faithfulness, retrieval quality, refusal adequacy, latency compliance, auditability, and human-review compatibility, and aggregates these into an institutional reliability index. It proposes a reliability-aware retrieval-augmented generation pipeline that integrates governed document ingestion, embedding generation, hybrid retrieval, reranking, evidence-aware generation, confidence-based refusal, human review, audit logging, and post-deployment monitoring. To support reproducibility, the paper compares four deployment configurations and provides an illustrative worked example of the reliability index. The contribution is a conceptual yet technically grounded deployment artifact that connects cloud computing, data science, and higher education governance; the architecture has not yet been empirically validated, and a protocol for future institutional pilots is specified.

Computers doi: 10.3390/computers15090563

Authors: Adarsh S. V. Nair Rathnakar Achary

Authentication in Web 3.0 faces a structural conflict. Systems that offer full anonymity leave no lawful way to identify a malicious actor. Systems built for accountability expose a persistent wallet address to blockchain-graph analysis, or fall back on centralized key recovery. Existing designs solve one side of this conflict at the cost of the other. This paper presents PEUAP-W3, a Privacy-Enhanced and User-centric Authentication Protocol. Its contribution is the integration of five established components into a single deployed and formally analyzed system. A Circom 2 circuit of 1579 Groth16 constraints proves four facts in a single 192-byte on-chain proof: knowledge of an opening of the session credential commitment, an SpO2 value inside an 85–100% band, single-use nonce binding, and HMAC integrity. Shamir (k = 2, n = 3) sharing distributes the identity payload across three independent relays. The coordinator reconstructs an identity only after a threshold vote has been recorded on chain. Credentials are issued as W3C Verifiable Credentials 2.0 in did:key form. Four Solidity contracts verify the proof on Ethereum Sepolia. Verification costs about 241,000 gas and takes roughly 3 ms. ProVerif and Scyther find no attack under the Dolev–Yao model. A concurrency sweep to 500 simultaneous requests completes 1191 requests with zero failures at about 15.4 requests per second. A behavioral gate screens commodity abuse as a supplementary control; it is not treated as a security boundary. Against a nine-property framework, PEUAP-W3 satisfies six properties. Three remain conditional and are not verified in the current deployment: biological-origin assurance and digital replay prevention, both of which need an attested sensor; and GDPR erasure equivalence. Here, formally verified refers to the protocol models and theorems, not to the complete deployed software.

Computers doi: 10.3390/computers15090562

Authors: Elisabetta Mazzullo Okan Bulut

Large language models (LLMs) are increasingly used to generate synthetic, human-like data for research and applied decision-making, including synthetic responses to self-report instruments such as surveys and questionnaires. As these synthetic datasets are folded into pipelines that inform scientific inference and downstream decision support, their trustworthiness becomes a practical concern. In human data, careless responding is a well-documented source of bias in self-report measures, yet no research has examined whether LLMs reproduce similar patterns of carelessness when generating synthetic responses. This study addresses that gap by comparing a large synthetic dataset generated by GPT-3.5-Turbo on a mental health questionnaire to a large human dataset, examining response diversity and multiple indicators of careless responding. GPT-generated scale responses were less variable than human responses, showing limited use of extreme response options, higher repetition, and reduced multivariate dispersion relative to a human-derived reference distribution. In contrast, the textual explanations accompanying the synthetic responses exhibited greater lexical variation, indicating that GPT’s linguistic ability did not translate into an ability to reproduce the heterogeneity of human psychometric response behavior. Carelessness-like patterns were also observed in the synthetic data, although conventional indices behaved differently across datasets and could not always be interpreted as direct evidence of human-like careless responding. These findings raise questions about the reliability of LLM-generated data for research and decision-making applications and point to the need for quality-assurance checks before trusting such data in downstream intelligent systems.

Computers doi: 10.3390/computers15090561

Authors: Ke Zheng Jincheng Jiang Zhixue Liang Yoo Youngjae Yufeng Wang

Lane detection is a core perception task for Advanced Driver Assistance Systems (ADAS) and autonomous driving. Current methods struggle to balance accuracy, model complexity and inference efficiency: high-precision models rely on heavy modules with excessive computation, while lightweight ones suffer from weak feature extraction and low precision. To alleviate this inherent trade-off, we propose YOLOv8n-LaneDG based on YOLOv8n-seg. We design a dual-path gated fusion block to strengthen lane features and an efficient upsampling convolution block to reduce computational overhead, and we further design a weighted continuity loss to preserve lane structural integrity. Evaluated on TuSimple, our method lifts mAP@0.5 from 74.3% to 95.2%. It outperforms mainstream lightweight models and matches heavy YOLOv8s-seg with far fewer parameters, delivering a high-precision, deployable lane detection solution for vehicle-end platforms.

Computers doi: 10.3390/computers15090560

Authors: Attila Kovari

Stable discrimination can coexist with unreliable probabilities, abstention policies, and uncertainty sets under distribution shift. We evaluate this mismatch with CRIT-AID, an executable multi-domain reliability-audit framework that separates model fitting, probability calibration, operating-rule calibration, and testing and then transports source-derived rules unchanged to target domains. Across four public tabular domains, stable discrimination did not imply stable probability quality, selective operating points, or conformal uncertainty. On identical ACS 2024 records, changing the income target definition left AUROC nearly unchanged, while ECE differed by 0.083; prevalence-intercept alignment reduced this difference to −0.004, showing that target semantics can alter probability reliability without materially changing ranking. Across 27 primary 90% conformal conditions, label-conditional calibration improved worst-class coverage in 18 but worsened it in 9 and usually increased prediction-set size. LightGBM sensitivity changed absolute discrimination without removing the mismatch among reliability dimensions. These findings show that discrimination alone is insufficient evidence for reliable AI decision support: audits should test the transportability of probability mappings, operating rules, class-specific validity, and uncertainty informativeness when deployment conditions or target meanings change.

Computers doi: 10.3390/computers15090559

Authors: Alexandru Răzvan Căciulescu Matei Bădănoiu Răzvan Rughiniș Eduard-Costin Dumistrăcel Dinu Țurcanu

Java middleware may expose Java Management Extensions (JMX) through Jolokia’s Hypertext Transfer Protocol (HTTP) bridge. In affected ActiveMQ deployments, reachable Log4j 2 configuration managed beans (MBeans) become write capabilities and, with compatible triggers, enable remote code execution (RCE). We ask: in a specified product/version state, which writes and triggers compose into RCE, and which operational guard first fails or remains unresolved? We synthesize two published case studies into an evidence-coded method. Four write and four trigger classes recover four observed chains, isolate one model-implied pairing, and reject dependency-level candidates through failed or unresolved guards. It distinguishes ActiveMQ paths from documentation-limited Apache James, WildFly, Apache Karaf, and Red Hat AMQ cases. Egress filtering and static-file ownership do not stop every observed chain. Observed (O), derived (D), and model-implied (H) labels separate findings from hypotheses. The contribution is a falsifiable, product- and version-scoped management-plane instrument, not a new attack-stage sequence.

Computers doi: 10.3390/computers15090558

Authors: Francesco Biondani Luigi Capogrosso Francesco Tosoni Nicola Dall’Ora Enrico Fraccaroli Franco Fummi

Recent advances in Digital Twin technology focus on visualization, operator training, and real-time machine simulation, aligning with the Industry 4.0 paradigm. However, the transition toward Industry 5.0 demands human-centric approaches that integrate workers not merely as observers but as active, monitorable parts of the system. Despite growing research interest, mature end-to-end methodologies for creating and deploying reliable Human Digital Twins (HDTs) in industrial environments are still lacking. This paper introduces Industrial Meta-Human (IMHU), an end-to-end human-centered Digital Twin methodology designed to bridge this gap. By spanning the entire lifecycle, from human modeling to production deployment, IMHU leverages Unreal Engine simulation to generate accurate human models and synthetic data, allowing safe replication of hazardous scenarios without disrupting ongoing operations. The methodology integrates Artificial Intelligence (AI) to enable real-time monitoring and support data-driven decision-making. Deployed on a fully operational production line, IMHU includes a system integration layer based on a Service-Oriented Architecture (SOA), enabling seamless interoperability with legacy Industry 4.0 infrastructures. Experimental results demonstrate the feasibility and confirm the effectiveness of real-time human-state tracking, and underscore its potential to advance scalable, human-centered Digital Twin systems for Industry 5.0.

Computers doi: 10.3390/computers15090557

Authors: Didar Dlshad Hamad Ameen Shahab Wahhab Kareem

This study proposes the Turtle Optimization Algorithm (TOA), a bio-inspired metaheuristic motivated by the long-distance navigation behavior of marine turtles. TOA integrates four main mechanisms: geomagnetic orientation modeled through sinusoidal modulation, ocean current drift, stamina-aware adaptive reference selection, and an Environmental Coordination Strategy (ECS). These mechanisms are jointly designed to balance exploration and exploitation while reducing premature convergence. The TOA was evaluated on 29 benchmark functions, including classical and CEC2019 benchmarks, and compared with established and recent metaheuristic algorithms. Friedman analysis revealed statistically significant differences among the compared methods (p < 0.05). The TOA achieved first place average ranks of 1.43 and 1.33 in two classical benchmark comparison groups. On CEC2019, TOA obtained average ranks of 1.60, 3.20, and 2.70, corresponding to first, second, and first place, respectively. The practical applicability of the TOA was further evaluated on three constrained engineering design problems—welded beam, speed reducer, and clutch brake design—where competitive solutions were obtained. Overall, the results demonstrate that the TOA provides a competitive and robust optimization framework across diverse benchmark and engineering problems.

Computers doi: 10.3390/computers15090556

Authors: Nurzhan Mukazhanov Zhibek Alibiyeva Ainur Akhmediyarova Bauyrzhan Ashirbekov Nurzhol Yerbolat Maksat Kanat

This study presents the development of an Artificial Intelligence (AI)-based legal assistant using the Retrieval-Augmented Generation (RAG) architecture to provide legal assistance to citizens of the Republic of Kazakhstan. The proposed solution is designed to generate accurate, evidence-based responses to user queries using the regulatory legal acts of the Republic of Kazakhstan as the primary source of information. A legal corpus comprising 101,000 legislative documents and court decisions, with approximately 77 million tokens in Kazakh and Russian, was constructed to support the retrieval component of the system. To identify the most effective semantic retrieval method, three multilingual embedding models—Multilingual-E5-Large, BGE-M3, and KazEmbed-V5—were evaluated for vector search. The experimental results showed retrieval accuracies of 87.6%, 76.8%, and 83.3%, respectively. The GPT-5.4 and Llama-4-Scout-17B-16E-Instruct large language models were used to generate legal reasoning and responses based on documents retrieved through semantic search. The quality of the generated responses was evaluated using two complementary approaches. First, legal experts assessed the factual correctness and legal validity of the answers. Second, automatic evaluation was performed using word-level F1, BLEU, ROUGE, and BERTScore-F1 metrics. Among all evaluated configurations, GPT-5.4 combined with Multilingual-E5-Large achieved the highest overall accuracy (88.5%), whereas Llama-4-Scout-17B-16E-Instruct combined with KazEmbed-V5 achieved an accuracy of 83.6%. Based on the proposed architecture and the selected semantic retrieval and language models, an AI legal assistant was developed and integrated into the “Adal Azamat” legal services platform providing users in Kazakhstan with practical access to AI-assisted legal consultation.

Computers doi: 10.3390/computers15090555

Authors: Liubov Smirnova Andrew Gekhtin Anna Maslovskaya

In silico studies of microbiological systems are essential for predicting and controlling the impact of external factors on bacterial communities. Quorum sensing represents one of the key mechanisms of bacterial communication, particularly in pathogenic bacteria, realized as a cell-density-dependent regulatory process governed by diffusible signaling molecules. The present study proposes a Physics-Informed Neural Network (PINN)-based computational framework for a spatially independent model of bacterial quorum sensing and population dynamics. The approach solves both the forward and inverse problems for a spatially independent model formalized by a system of nonlinear ordinary differential equations. The forward problem is solved numerically by reconstructing the dynamics of three key characteristics: signaling molecule concentration, degrading enzyme concentration, and bacterial biomass density. The obtained PINN solutions are compared with numerical solutions computed using the Radau IIA implicit Runge–Kutta method. The inverse problem capability is evaluated by recovering system parameters that are difficult to measure directly in experimental settings. The framework is implemented using the DeepXDE library with a PyTorch backend, employing hard constraints for initial conditions, singularity-avoiding loss reformulations, and a multi-stage Adam–L-BFGS optimization strategy. Validation is performed on a Monod chemostat benchmark and the Pseudomonas putida IsoF quorum sensing regulatory network. The proposed PINN-based framework extends the applied mathematical toolkit for in silico studies of microbial systems, enabling accurate reconstruction of emergent population dynamics and robust inference of regulatory parameters that are inaccessible to direct experimental measurement.

Computers doi: 10.3390/computers15090554

Authors: Georgios Yfantidis Panagiotis D. Michailidis

Social media has become an important tool for understanding visitor experiences in museums and cultural heritage sites. This literature review identifies, organizes, and thematically synthesizes existing studies on museum visitor experience based on social media data. It examines 41 studies retrieved from Scopus and Web of Science and published between 2017 and 2026. Furthermore, the review examines the selected studies across six dimensions: social media platforms, types of user-generated data, the role of digital interactions, the museums and cultural heritage sites studied, the analytical methodologies applied, and the main findings on visitor experience. The findings indicate that TripAdvisor is the most frequently used platform for collecting textual reviews and star ratings, whereas Instagram and Flickr are mainly used for visual and spatial data. Most studies rely on computational methods, often combined with quantitative techniques, while qualitative approaches are used less frequently. The identified methods include content analysis, statistical analysis, sentiment analysis, topic modeling, machine learning, image analysis, and spatial analysis. Across the reviewed studies, visitor experience is examined as a multidimensional phenomenon encompassing emotions, service quality, authenticity, historical connection, aesthetics, education, and social participation. Finally, the review identifies recurring themes across the dimensions and synthesizes them into broader research streams. These are brought together in an integrative synthesis framework that organizes existing research, highlights research gaps, and outlines directions for future studies.

Computers doi: 10.3390/computers15090553

Authors: Oswaldo A. Peña Rojas German Sanchez-Torres John W. Branch-Bedoya

Degraded tax forms and other legal–administrative documents require restoration methods that impose strict constraints on content fidelity. This paper presents a structure-aware restoration framework for degraded tax documents that combines geometric normalization, canonical template guidance, and supervised image restoration within a common alignment space. Documents are first mapped to a canonical layout through homography estimation based on Scale-Invariant Feature Transform (SIFT) and Random Sample Consensus (RANSAC), using the canonical visual template as the reference. Restoration is then performed using template priors and masked constraints designed to preserve the fixed document structure while recovering variable content. Under a fixed-budget comparative protocol, U-Net with template priors achieved the highest visual and structural quality, whereas the proposed hybrid model obtained the best functional OCR performance on synthetic data, reaching a Character Error Rate (CER) of 0.1091 and a Word Error Rate (WER) of 0.3495. As a complementary evaluation on 37 real-world documents with human-generated textual ground truth, restoration increased full-page word coverage across all three OCR engines evaluated, yielding absolute improvements ranging from 4.15 to 18.05 percentage points.

Computers doi: 10.3390/computers15090552

Authors: Assem Shormakova Madina Mansurova Beibitkhan Yerkegul Marek Milosz

This article investigates offline recognition of handwritten Kazakh text in the Latin script using a convolutional recurrent neural network. The relevance of the study is determined by the transition of the Kazakh language to the Latin alphabet and the need to automate the processing of handwritten documents. The proposed model consists of a convolutional neural network feature extractor, two bidirectional long short-term memory layers, and a Connectionist Temporal Classification decoder. The convolutional layers extract visual features from word images, the bidirectional recurrent layers model the sequential relationships between characters, and CTC enables end-to-end training without explicit character-level segmentation. A specialized dataset named KazEsim, containing 20,000 handwritten Kazakh name images, was created and divided into writer-independent training, validation, and test subsets. Experimental results showed a character accuracy rate of 96.5% and a word accuracy rate of 92.3%. Compared with a conventional CNN baseline, the proposed CRNN model improved character accuracy by 6.1 percentage points and word accuracy by 9.2 percentage points. The proposed model also outperformed the fine-tuned TrOCR-small comparative baseline while requiring fewer parameters and lower inference latency. These findings demonstrate the effectiveness of CNN–BiLSTM–CTC sequence modeling for offline recognition of handwritten Kazakh words in the Latin script.

Computers doi: 10.3390/computers15090551

Authors: Tian Wang Masooda Bashir

AI-powered conversational agents are becoming part of the everyday Internet information ecosystem, reshaping how users seek, interpret, and act on health-related information outside clinical encounters. As large language model (LLM)-based chatbots are increasingly used as on-demand digital health information tools, understanding how users perceive their credibility, usefulness, and limitations is essential for the responsible design of future Internet-based health services. This mixed-method survey study examined how general adults evaluated healthcare-related question–answer pairs provided by physicians and generated by AI chatbots. A sample of U.S.-based adults recruited through Prolific (N = 62) rated each answer on clarity, usefulness, appropriateness of detail, trustworthiness, and perceived evidence, and provided open-ended explanations of their judgments. Primary mixed-effects analyses showed that both ChatGPT- and Claude-generated responses received higher overall participant ratings than physician-provided responses, although the estimated difference was substantially larger for Claude (ChatGPT–physician estimate = 0.250, 95% CI [0.135, 0.364]; Claude–physician estimate = 0.825, 95% CI [0.710, 0.939]). ChatGPT received higher ratings on four of the five dimensions but not on clarity, whereas Claude received higher ratings across all five dimensions. However, physician, ChatGPT, and Claude responses were always presented first, second, and third, respectively. Response source was therefore confounded with presentation position, and the observed differences cannot be attributed exclusively to source. The responses were also not matched for length or format. Qualitative findings showed that participants valued detailed, specific, and evidence-like explanations. Participants also expressed concerns about hallucination, privacy, over-reliance, and the need for clinician verification. These findings suggest that LLM-based chatbots may be perceived as useful supplemental information tools within future Internet health ecosystems, but their deployment should include safeguards that support transparency, verification, and appropriate reliance.

Computers doi: 10.3390/computers15080550

Authors: Zhijiang Li Chuan Ding

Power line component and insulator defect detection is an important task in intelligent transmission line inspection. However, UAV-based inspection images often contain small, densely distributed targets under complex backgrounds, making it difficult for existing detectors to balance accuracy and model compactness. To address these challenges, this paper proposes a lightweight YOLOv8-based detector named DSCMamba-TAD-YOLOv8. First, depthwise separable convolutions are introduced into the Neck to reduce parameters and computational cost. Second, DSCMambaNet replaces the original C2f module to enhance multi-scale feature representation by combining lightweight local feature extraction and cross-region contextual modeling. An embedded CBAM component is further integrated inside DSCMambaNet to strengthen informative channel responses and spatial regions. Finally, a Task-Aware Dynamic Detection Head, named TADetect, improves head adaptability through scale-aware and task-aware feature modulation. Experiments on the InsPLAD-det dataset show that DSCMamba-TAD-YOLOv8 achieves 91.86% Precision, 88.02% Recall, 91.83% mAP@0.5, and 74.82% mAP@0.5:0.95. Compared with YOLOv8n, the proposed model improves Precision, mAP@0.5, and mAP@0.5:0.95 by 4.09, 2.43, and 4.46 percentage points, respectively, while maintaining a comparable Recall level with a slight increase from 87.04% to 88.02%. Meanwhile, Params decrease from 3.209 M to 2.702 M and GFLOPs from 8.2 to 7.5. On the revised TPL-SOD held-out test subset, the proposed model improves Precision from 86.20% to 88.16%, mAP@0.5 from 87.09% to 88.81%, and mAP@0.5:0.95 from 68.44% to 70.13%, while Recall remains stable and slightly increases from 91.75% to 92.33%. These results demonstrate that DSCMamba-TAD-YOLOv8 improves detection accuracy and localization quality while maintaining a compact structure and stable recall performance.

Computers doi: 10.3390/computers15080549

Authors: Weiqiang Chen Zhiyao Zhao Haisheng Li Jiping Xu Chongxuan Liu Xin Zhang

Fruit and vegetable supply chains generate heterogeneous data across production, storage, logistics, and sales, creating challenges for trusted data sharing, privacy protection, and real-time traceability across distributed supply-chain information systems. Conventional single-chain blockchains suffer from limited scalability, data redundancy, and low retrieval efficiency, making them inadequate for high-frequency full-process information management. This study proposes a multi-chain blockchain framework for trusted full-process information management of fruit and vegetable supply chains. The framework integrates traceability, enterprise, notary, and regulatory chains to support hierarchical data management and privacy isolation. A reputation-based notary node election mechanism and a threshold-signature scheme based on Shamir secret sharing are designed to enhance cross-chain security and distributed regulatory consensus. To improve retrieval efficiency, a Cuckoo-Augmented Merkle Tree (CMerkle) and a skip-list-based block index are developed. Simulation results show that all malicious nodes were restricted by the 19th round, signature aggregation required 70.16 ms in a 500-node setting, and CMerkle achieved retrieval speedups of 14.7 and 153 times at data scales of 500 and 10,000 records, respectively. The framework supports trusted data governance, real-time traceability, privacy-preserving sharing, and regulatory decision support in blockchain-enabled supply-chain information systems.

Computers doi: 10.3390/computers15080547

Authors: Tudor-Mihai David Mihai Udrescu

Poisoning attacks present a significant challenge for federated learning (FL), particularly when clients operate with Non-IID data and updates are transmitted through multi-tier aggregators. This paper introduces FLVaccin, a hierarchical personalized FL framework structured as an unbalanced tree. In this framework, each non-root node both hosts local clients and aggregates shared MobileNetV2 features (FedPer), whereas the root node does not possess local data. The proposed defense integrates node-level CIFAR-100 vaccination, which calibrates depth- and round-adaptive tolerances, with per-client trend quarantine and root backbone rejection. Experimental results on CIFAR-10 with 100 clients, 25 aggregators, and Dirichlet Non-IID partitioning (α=0.5) demonstrate that the clean baseline achieves 79.9% accuracy. In contrast, unconstrained mixed attacks reduce performance to near-chance levels (20.2% k-fold). When vaccination and quarantine are enabled, 535 attack events still result in a 76.5% ± 0.4% k-fold accuracy (77.3% test), remaining within 2.6 percentage points of the clean model. These findings indicate that tree-aware, vaccination-calibrated monitoring can maintain model utility under persistent multi-vector poisoning without the need to share raw data.

Computers doi: 10.3390/computers15080548

Authors: Leonidas Boutsikaris George Katrilakas Athanasios Tsadiras Symeon Samaras Christina Topalidou

Graph neural networks remain challenging to apply to traffic forecasting, often failing to consistently outperform simple baselines. We observe that the strongest baseline is regime-dependent: last-value persistence dominates at short horizons on smooth freeway data, while time-of-day climatology prevails at longer horizons and on bursty arterial networks. Rather than treating this as a limitation, we propose a learnable approach that automatically selects the optimal baseline. Our method anchors predictions to a learned, per-horizon convex blend of persistence and climatology, introducing only twelve scalar parameters. This learned anchor recovers whichever baseline is locally most effective, allowing the model to focus on capturing residual variations that neither baseline captures. We evaluate on six public benchmarks (METR-LA, PEMS-BAY, PEMS03/04/07/08) spanning traffic speed and flow data under standard 70/10/20 chronological splits with masked evaluation metrics and holiday-aware climatology. Our anchored temporal models consistently beat both baseline methods on the 12-step average across all datasets, and outperform at every horizon on five of the six benchmarks. When integrated into two strong architectures (STID and Graph WaveNet), the anchor yields substantial gains at long horizons where climatology is most informative. Notably, within our lightweight framework, learned spatial graph components do not improve accuracy and can slightly degrade performance, a finding we analyze and discuss.

Computers doi: 10.3390/computers15080546

Authors: Eman Alsaidi Eman Omar Basela Hasan

In this research, we propose a graph-based approach to classify ECG signals. We model the ECG signal as a time-varying graph and examine its dynamics at two levels: intra-beat and inter-beat. The proposed framework employs a two-level graph representation of the ECG. At the intra-beat level, each beat is modeled as a graph in which the P, Q, R, S, and T waves serve as nodes, and temporal distance information is incorporated into the node feature vector. At the inter-beat level, the ECG signal is represented as a graph of beats. We investigated two node representations: learnable MLP embeddings derived from handcrafted features and reduced features obtained via PCA. Unlike conventional temporal GCNs, graph attention networks (GATs), and transformers, which predominantly rely on single-level representations of ECG signals, the proposed approach builds hierarchical graph structures directly from ECG signals. The dual model was trained on 1024 ECG segments from the PTB Diagnostic ECG Database (PTBDB) and tested on 398 ECG segments from the MIT-BIH Arrhythmia Database to evaluate its performance. The results indicate that the best-performing configuration, which employs PCA-reduced node features, outperforms the other configurations tested. The model achieves a mean accuracy of 97.67% and a mean F1-score of 97.18% over five runs.

Computers doi: 10.3390/computers15080545

Authors: Zhao Shen Rulong He Xiao Zhang

We examine whether simulated attack-action evidence can inform a defender that they must commit before an attacker’s type is known. We formulate a five-stage Bayesian Stackelberg security game with five stage-specific actions per player. A Monte Carlo predictor produces type-conditioned action likelihoods on an enterprise graph, while prediction confidence weights the next Bayesian update. The defender strategy is computed by exact follower-response enumeration and linear programming. Evaluation used a simulated 10-node enterprise network, 30 paired trials, bootstrap confidence intervals, and Holm-adjusted Wilcoxon tests. Against a fixed-prior Bayesian Strong Stackelberg Equilibrium, mean gross defense utility increased from 2.141 to 2.197. Mean attack success decreased from 0.691 to 0.686. The paired differences remained significant after multiplicity correction. Outcomes did not differ significantly from an equilibrium updated with coarse reference likelihoods, and the simulation cost reduced net utility by 0.08. Maximum follower regret and constraint violation remained at the specified numerical tolerance. Runtime remained near 0.39 s across networks of 10–100 nodes. An action/type experiment showed rapid growth as follower-response profiles increased. Exact commitment and sequential updating were feasible in the abstraction; simulation was not automatically cost-effective when a usable reference model was available.

Computers doi: 10.3390/computers15080544

Authors: Borivoj Bogdanović Siniša Nikolić

Domain-specific information systems often maintain their data model, rule base, and application infrastructure as separate artifacts, complicating maintenance and pre-deployment verification. This study investigates whether these artifacts can be unified in a verifiable ontology-to-code pipeline without changing the expected classifications. The proposed Model-Driven Architecture uses the Business Application Builder framework and a Web Ontology Language 2 Description Logic ontology to represent domain structure, classification rules, and generation metadata. HermiT verifies consistency, satisfiability, and subsumption under open-world semantics before code generation. The generator produces persistence, business-logic, data-transfer, and presentation layers, while the generated Java application evaluates stored records under closed-world semantics and resolves overlapping categories using ontology-declared priorities. In a Serbian research-evaluation case study, the generated system reproduced the M30 and M33 classifications of an established Jess implementation. An internal secondary experiment generated and executed a prenatal-diagnosis application; all six runtime classifications matched the HermiT entailments and expected outcomes. The public artifact independently reproduces the ontology-level experiments but excludes the proprietary generator and generated source code. The results support the feasibility of ontology-driven generation for static-classification systems, whereas arithmetic risk computation and temporal event processing remain better suited to complementary procedural technologies. No performance superiority is claimed.

Computers doi: 10.3390/computers15080543

Authors: Ali Fadhel Athab Hadi Seyedarabi Reza Afrouzian

In this research, a new approach is proposed to accurately predict indoor three-dimensional localization based on Received Signal Strength (RSS) values. This work uses WiFi beacons to collect time-series RSS data, preprocess it, and feed it to the proposed model. The proposed model presents a novel architecture based on 2D convolutional neural networks, and this model employs a multitask learning approach. Hence, the model simultaneously has a classifier for floor classification and a regressor for estimating X and Y coordinates, and tries to perform accurate indoor localization even in environments with furniture and other obstacles. The proposed CNN-based model efficiently utilizes RSSI data, achieving 99.0% floor classification accuracy and 6.5 m in terms of Euclidean distance error based on coordinate estimation on the UJIIndoorLoc dataset. On the other hand, the validation results using Tampere datasets were ‘distance error (m) = 3.7868’ form localization and ‘accuracy = 98.99%’ for floor classification. Comprehensive preprocessing significantly enhances localization accuracy.

Computers doi: 10.3390/computers15080542

Authors: Ahmed M. Elalfy Gamal A. Ebrahim Marvy Badr Monir Mansour

The deployment of deep-learning vulnerability detectors in production remains difficult. Models are large, false-positive rates are high, output is opaque, and a persistent gap separates benchmark performance from real-world utility. The objective of this work is to close part of that gap by combining a learned detector with interpretable rules so that accuracy, efficiency, and actionability are achieved together. A hybrid framework is therefore presented in which a lightweight edge-conditioned GNN of 71,810 parameters, named FastVulnGNN, trained in 96.2 s on a single CPU core, is paired with rule-based heuristic detection for six C/C++ vulnerability classes, namely buffer overflows, format-string defects, null-pointer dereferences, double-free errors, integer overflows, and race conditions. On the MegaVul dataset, an accuracy of 71.1%, an F1 score of 0.70, and an AUC-ROC of 0.77 are obtained by the GNN component. On a production codebase of 499 files and 312,758 lines of code, the full hybrid scan completes in 5.5 s, which corresponds to about 57,000 lines per second, without any GPU hardware. Per-file risk tiers and pattern-level explanations are produced, and these are suitable for continuous-integration use. The significance of this work lies in demonstrating that a deployable, explainable detector can be assembled from compact components, and an edge-type ablation study, a cross-dataset evaluation, and a per-vulnerability analysis are reported to characterize the approach.

Computers doi: 10.3390/computers15080541

Authors: Pingyuan Zhang Limin Wang

Rejection samplings are the essential building blocks to design lattice-based digital signatures under the Fiat–Shamir paradigm. Up to now most of them have been built by Gaussian samplings or uniform samplings. Gaussian-based rejection sampling signatures such as the BLISS scheme have very short signature sizes but are vulnerable to timing attacks, whereas the uniform-based rejection sampling signatures such as ML-DSA are allowed to be simply implemented but have larger signature sizes. This work intends to use a new probability distribution, rather than Gaussian or uniform distributions, to build the rejection sampling in Fiat–Shamir signatures, and aims to achieve short signature sizes while avoiding the cost of Gaussian sampling. To this end, we choose centered binomial distribution as a replacement, and build a new and compact rejection sampling that has the properties of both high-precision sampling and semi-uniform operation. As an application, we combine this rejection sampling with Lyubashevsky’s signature scheme, then propose the first lattice-based Fiat–Shamir signature scheme from centered binomial distribution. The proposed scheme not only avoids Gaussian sampling, but also is very efficient in terms of the signature sizes.

Computers doi: 10.3390/computers15080540

Authors: Bingchi Sun Haibin Zheng Jingjing Jin

In Industrial Internet of Things (IIoT) deployments, mobile edge computing (MEC) offloads computation-intensive tasks, a constrained biobjective problem trading time delay against energy consumption (MTOP). We show that this benchmark is exactly separable across micro base-station (MiBS) regions: its delay and energy objectives are additive over regions, and the only coupling, intra-cell interference, stays within a region. Exploiting this, we propose CR-MTMEMTO-D, a structure-aware decomposition multitasking method that treats each region as an independent subtask, solves it with a feasibility-repaired NSGA-II, and reconstructs the global feasible Pareto front as the non-dominated subset of the Minkowski sum of the regional fronts, an exact composition that adds no global evaluations. Across 12 instances (45–432 variables, 20 seeds), it attains the best hypervolume and IGD on every instance (mean HV 0.9340 vs. 0.8021 for a plain NSGA-II baseline; average rank 1.00), with the margin widening as the problem scales, and it is unchanged under total-evaluation matching because every evaluation is a regional main task. A feasibility-priority acceptance gate keeps the population fully feasible. Under matched budgets, a prior cheap-task pool with bandit-controlled transfer adds no significant gain, which motivates the structural approach.

Computers doi: 10.3390/computers15080539

Authors: Wellington Kanyongo Mampilo Phahlane

AI-generated synthetic media are evolving from tools of digital manipulation into a practical antidote for persistent challenges in digital health implementation. However, the computational capabilities that characterise these technologies, their applications and downstream health-related benefits remain fragmented and insufficiently synthesised. This systematic review identified the computational capabilities that characterise AI-generated synthetic media in health, examined their applications and benefits, and developed an integrative framework linking these domains. Twenty-four studies published between 2021 and 31 May 2026 were included. Methodological quality was assessed using the Mixed Methods Appraisal Tool (MMAT) and findings were synthesised through thematic analysis. The synthesis revealed an integrated set of capabilities spanning photorealistic medical-image generation, modality-specific synthesis of clinical images and physiological signals, synthetic non-image health-data creation, preservation of statistical distributions, temporal patterns and clinical relationships, generation of diverse, novel and non-memorised samples and controlled transformation of medical and audiovisual content. Privacy-oriented synthesis and deepfake detection emerged as distinct components supporting privacy-conscious data use, clinical verification and healthcare safety. These demonstrated capabilities were linked to empirically evaluated and indicated applications, including data augmentation, AI model training, diagnostic model development, privacy-oriented health-data sharing, medical education, patient-facing communication, therapeutic support, clinical safety, health-system analytics and planning. The resulting Computational Capability–Application–Benefit (CAB) Framework conceptualises synthetic media as an evidence-graded pathway distinguishing demonstrated computational capabilities, evaluated health-related applications and reported, indicated or potential downstream benefits requiring further validation. AI-generated synthetic media, therefore, represent an emerging computational infrastructure with potential to support safer, privacy-conscious, adaptive and data-intensive healthcare.

Computers doi: 10.3390/computers15080538

Authors: Vítor J. Sá Paulo Veloso Gomes João Donga Rosalina Babo António Marques

The increasing complexity of healthcare data ecosystems demands educational technologies capable of supporting data-intensive learning through advanced analytics, immersive interfaces, and learning analytics. This paper presents a cloud-based reference architecture and a prospective evaluation protocol for integrating Business Intelligence (BI), Extended Reality (XR), and learning analytics in health data science education. The proposed architecture is informed by a systematic literature review conducted according to the PRISMA 2020 guidelines, which screened 613 records retrieved from four databases and retained 56 studies for qualitative synthesis. The review indicates that, although BI and XR technologies have independently been associated with educational benefits, empirical evidence supporting integrated educational architectures combining BI, XR, and learning analytics remains limited, particularly in health data science education. Based on these findings, the paper specifies a layered reference architecture comprising a cloud analytics engine, an immersive visualization engine, an interoperability layer, and a learning analytics pipeline designed to support adaptive and AI-assisted educational services during subsequent implementation phases. The reference architecture is partially instantiated within the curricular unit Health Data Analysis and Visualization of the Digital Health programme at the Polytechnic University of Porto, where the BI and XR components are currently deployed and used within the course, while the interoperability middleware, learning analytics infrastructure, and AI-assisted services remain under development or are specified as architectural capabilities. To support future empirical validation, the paper also defines a comprehensive prospective evaluation protocol comprising predefined outcomes, established instruments with published psychometric properties, together with an expert-developed health data literacy assessment undergoing content validation, research hypotheses, power analysis, a statistical analysis plan, and ethical and data-governance provisions. The manuscript makes four principal research contributions: (i) a cloud-based reference architecture for BI–XR integration, (ii) a computational learning analytics pipeline specification, (iii) an interoperable system design for health data science education, and (iv) a prospective evaluation protocol to guide the future validation of the proposed reference architecture.

Computers doi: 10.3390/computers15080537

Authors: Mingwei Cao Zilong Wang Ning Li Haifeng Zhao

In the field of 3D reconstruction for photovoltaic scenes, current texture mapping techniques frequently encounter significant texture segmentation and apparent joins in uniform plane regions, such as solar panels, because they lack geometric structural assumptions. To tackle these challenges, we introduce a new texture-mapping strategy for 3D solar panel scene reconstruction that focuses on planar structure awareness. We term the proposed method PSATM. Initially, we suggest a global constraint and a local refinement process to incorporate clear geometric structure details. This process automatically detects and labels planar regions through a region-growing approach. Next, we integrate a planar structure-aware module into the smoothness term of the Markov Random Field (MRF) energy function. This module uses dihedral angles and plane membership to adjust label transition costs, enhancing texture coherence within planar regions and maintaining smooth transitions at genuine geometric breaks. Furthermore, we establish a boundary treatment technique relying on local geometric support. This method combines area-based weighting and normal consistency to modify erroneous labels, successfully removing small remnants and smoothing texture edges. We tested the proposed PSATM with texture patch counts and visual quality measures on actual solar panel scenes. The results indicate that the proposed PSATM considerably reduces texture segmentation errors and improves texture flow and overall visual quality compared to the existing method.

Computers doi: 10.3390/computers15080536

Authors: Usman Mohyud din Chaudhary Humaira Arshad Muhammad Ismail Mohmand Erum Ashraf Waheed Ali H. M. Ghanem

Cross-chain bridges represent one of the most damaging attack surfaces in decentralized finance, with major exploits (e.g., Ronin, Wormhole, Nomad, Multichain) arising not from broken signature schemes but from failures in proof verification, replay protection, and signer-set management, gaps that conventional threshold-signature multisignature wallets do not address. This study presents an incident-aware multisignature architecture combining three on-chain predicates—block-height freshness windows, epoch-bound signer sets, and Merkle inclusion-proof verification—with a non-authoritative off-chain LightGBM classifier that generates SHAP-attributed risk explanations to support governance actions such as pausing, vetoing, or rotating signers, without directly blocking or approving execution. The framework was evaluated on a simulated benchmark of 78,600 Ethereum testnet transactions containing six injected anomaly classes (gas spikes, nonce jitter, malformed call data, stale intents, proof-delivery delays, and epoch-rotation replays). The LightGBM advisor achieved ROC-AUC 0.92 (95% CI [0.906, 0.926]) and F1 0.73 ([0.712, 0.749]), outperforming five baselines—logistic regression, Random Forest, XGBoost, isolation forest, and a rule-based detector—with the highest F1 (0.731) and PR-AUC (0.799), while the rule-based detector, which by construction covers only the anomaly classes addressed by the deterministic predicates, attained F1 0.282. Differences were statistically significant except for the LightGBM–XGBoost PR-AUC comparison. The deterministic layer itself is verified through 28 property-level contract tests covering all seven modeled attack objectives, with measured per-function gas costs (execute_Intent: 118,756 gas, of which 28,432 gas is Merkle-proof verification). Within this controlled setting, the results indicate that a machine learning advisor can extend anomaly-prioritization coverage beyond the scope of the deterministic predicates while leaving execution control fully deterministic. This work is presented as a controlled proof of concept: the reported metrics quantify recovery of scripted injection patterns, and validation against real-world exploit traces remains future work.

Computers doi: 10.3390/computers15080535

Authors: Frédéric Pattyn Yannick Dillen Peter Goetz

Software startups operate in environments characterized by rapid change, high uncertainty, and limited resources, resulting in high failure rates and challenges such as premature scaling and cash flow mismanagement. Prior research on pre-MVP validation has largely measured activity by volume rather than by the strength of evidence produced, leaving open whether evidence type, rather than quantity, is associated with startup performance. This study addresses that gap by investigating how early-stage software startups validate their initial idea before building their first Minimum Viable Product (MVP). Through 29 semi-structured interviews with founders from 16 software startups, pre-MVP validation activities were extracted and inductively coded into a six-level Validation Canvas spanning three validation stages identified in the literature: problem validation, problem-solution fit, and product-market fit. Startup performance was assessed through a composite ranking across funding, revenue, profitability, and runway indicators, and validation activities were analyzed thematically to derive the six evidence levels. No clear relationship was observed between the number of validation events and startup performance. Instead, stronger-performing startups tended to reach higher levels of evidence—particularly securing contingent investment commitments (Level 5) or paying customers (Level 6) before full MVP development. Level 6—paying customers before the full product exists—is identified as the strongest form of pre-MVP market evidence, as it directly validates willingness-to-pay without relying on investor confidence. In this study, product-market fit is operationalised as demonstrated commercial viability through external financial commitments rather than interest signals or free sign-ups alone. Based on these exploratory findings, the study proposes the Hierarchy of Validation: a staged, bidirectional process model in which bottom-up traversal from informal interest signals (L1) toward paying customers (L6) emerged as the primary pattern among stronger-performing startups. A top-down direction, in which experienced founders begin at higher evidence levels and work downward, is proposed as a hypothesis for future research. To our knowledge, this is among the first accounts of pre-MVP validation that differentiates strength of evidence rather than volume of activity, contributing the Hierarchy of Validation as an original, exploratory framework for early-stage software startups. These findings remain exploratory and require validation in larger and more diverse samples.

Computers doi: 10.3390/computers15080534

Authors: Mohiuddin Mehedi Abdul Aziz Kabir Khandakar Rabbi Ahmed Furqaan Mujtahid Sakib Salam Jamee Md Nayem Rahman

Despite the benefits associated with the use of smart meters in advanced metering infrastructure, the widespread deployment of such meters has introduced vulnerabilities that leave power systems susceptible to stealthy false data injection attacks which cannot be detected by conventional methods. In this paper, we present a framework that combines a bidirectional long short-term memory network with an attention mechanism and a blockchain integrity layer to provide secure anomaly detection. We utilize the Smart Meter Electricity Consumption Dataset, augmented with synthetically injected anomalies, to detect abnormal consumption behavior. Experimental results demonstrate that the proposed hybrid model achieves up to 96.23% accuracy, 99.88% precision, 92.58% recall, and 96.09% F1-score, outperforming eXtreme Gradient Boosting (XGBoost), Isolation Forest, and Random Forest baselines. Confusion matrix analysis confirms minimal false positives and strong detection capability for stealthy attacks. The blockchain layer ensures immutability and trustworthiness of detection results through cryptographic hashing and consensus mechanisms with negligible overhead. The proposed framework offers a scalable, secure, and interpretable solution for defending smart grid infrastructures against complex cyberattacks.

Computers doi: 10.3390/computers15080533

Authors: Wassila Korichi Akram Zine Eddine Boukhamla Nadjet Azzaoui Mohamed Chahine Ghanem

Flying Ad Hoc Networks let unmanned aerial vehicles communicate directly, without relying on fixed ground infrastructure, in time-critical settings such as disaster response, search and rescue, border surveillance, precision agriculture, and military reconnaissance. UAV mobility, however, causes frequent topology changes and intermittent connectivity, so nodes typically store data locally and replicate it across the network to keep it available. The resulting challenge is consistency: independently evolving copies must be reconciled without a central coordinator, and strong consistency is not realistic in a network this prone to partitioning. We address this with the Decentralized Mobile Reconciliation Protocol (DMRP), which provides eventual consistency among UAV nodes with no external coordination, with convergence formally guaranteed whenever the swarm’s synchronisation graph is eventually connected. DMRP combines immediate local validation and convergence guarantees grounded in conflict-free replicated data type properties; hysteresis-based memory management with dual thresholds to cap journal storage overhead; adaptive delta or full-state synchronisation based on receiver lag; and epidemic propagation for transitive update dissemination. Energy efficiency guided the design throughout, through wireless broadcast and the avoidance of redundant transmissions. DMRP was implemented and evaluated through extensive OMNeT++/INET simulations of three-dimensional FANET scenarios. Results demonstrate that the protocol maintains a strictly bounded reconciliation journal, whereas the reference δ-CRDT log grows without bound, reducing reconciliation-journal storage by up to 75% at the largest workload evaluated, while achieving near-complete consistency after node isolation and network partitioning.

Computers doi: 10.3390/computers15080532

Authors: Sunghae Jun

Patent document–keyword count data are typically high-dimensional, sparse, and dominated by zero entries, which makes it difficult to simultaneously reconstruct keyword frequencies and identify meaningful technological structures. This study proposes a machine learning approach to latent structure learning for zero-inflated patent keyword count data. The proposed zero-gated latent factor model (ZG-LFM) combines nonnegative matrix factorization (NMF) with keyword-specific logistic occurrence models. NMF is used to extract interpretable document–factor and factor–keyword representations, while the occurrence gate estimates the probability that each keyword appears in a given patent document. The method was evaluated in an initial domain-specific case study using a document–keyword matrix constructed from 9434 quantum computing patent documents and 175 keywords, of which 87.60% of the entries were zero. Predictive performance was assessed using root mean squared error, mean absolute error, and the area under the receiver operating characteristic curve across different numbers of latent factors. The experimental results showed that NMF provided more accurate keyword count reconstruction, whereas the proposed model consistently achieved better discrimination between zero and nonzero keyword entries. These findings indicate that latent count reconstruction and keyword occurrence modeling provide complementary information for analyzing sparse patent data. The learned latent factors further revealed coherent quantum computing subdomains, including hybrid quantum–classical execution, quantum machine learning, quantum state measurement and error analysis, quantum cryptography, superconducting chips, quantum circuits, optical control, qubit devices, and optimization algorithms. The proposed framework therefore provides interpretable latent technology structures while improving the identification of keyword occurrence patterns in zero-inflated patent data. These findings demonstrate the feasibility of the framework within the analyzed quantum computing corpus; its generalizability across other technological domains remains to be evaluated.

Computers doi: 10.3390/computers15080531

Authors: Özcan Dimez Fatih Cogen

Mobile-network operators must interpret spatial anomalies in aggregated cell-level telemetry and decide whether, where, and how to respond. This paper presents a privacy-conscious, intrusion detection system (IDS)-oriented triage and response architecture that consumes cell-level anomaly signatures and couples spatial reconstruction, short-horizon forecasting, origin inference, self-resolution and remaining-time estimation, adaptive gating, ETA-aware team selection, conservative redeployment, explanation, and audit logging. It is a downstream spatial-attribution and response-orchestration layer, not a packet- or flow-level attack detector. The evaluated configuration uses transparent deterministic, heuristic, and optimization-based procedures and synthetic aggregated signatures without subscriber identifiers; aggregation is treated as data minimization, not a formal privacy guarantee. Across 20 paired synthetic scenarios, the full policy reduced conditional mean response time from 37.58 to 22.86 min, total travel from 576.0 to 273.5 min, and coverage ETA from 32.28 to 26.76 min, while on-time service increased from 54.0% to 60.0%. These benefits were accompanied by lower persistent-incident coverage (91.1% to 72.1%) and a higher miss rate (8.9% to 27.9%). The inverse-origin configuration showed no repeated localization-error advantage, and conservative redeployment had only a marginal average effect. The results therefore demonstrate a configurable downstream triage trade-off under controlled synthetic conditions, not attack-classification accuracy, adversarial robustness, formal privacy, or deployment readiness.

Computers doi: 10.3390/computers15080530

Authors: Gulnara Kaziyeva Gulzira Abdikerimova Anargul Bekenova Saule Zhumagulovа Gulden Murzabekova Ainur Shekerbek Balganym Kosherova Shynar Turmaganbetova Assem Aubakirova

This study proposes a hybrid computational model for diagnosing such biases using groundwater observation data across Kazakhstan. The analytical dataset included 2402 georeferenced observations, including 492 natural springs from OpenStreetMap (OSM), 109 boreholes from OSM, and 1801 spatially filtered pseudo-absence observations. Springs and boreholes together formed 601 positive groundwater observations, while pseudo-absence samples represented a spatially filtered background level rather than confirmed groundwater absence. Each observation was characterized by 89 environmental predictors extracted from Google Earth Engine. The proposed hybrid observation source bias index (HOSBI) combines a normalized robust effect size based on the median absolute value of the Cliff delta, multivariate distribution divergence quantified using RBF-MMD, and spatially confirmed source distinctiveness. These components were assigned fixed weights of 0.40, 0.35, and 0.25 to emphasize statistical and distributional data while maintaining spatial validation. Spatial cross-validation achieved a balanced accuracy of 0.855 for distinguishing OSM sources from OSM wells and 0.846 for separating positive observations from background pseudo-absences. Climate showed the strongest source-related bias (HOSBI = 0.923), while Sentinel-1 SAR contributed the most to the contrast between positive and background data (HOSBI = 0.923). The proposed framework provides an interpretable and replicable preliminary assessment of source bias in heterogeneous geospatial datasets.

Computers doi: 10.3390/computers15080529

Authors: Attila Körei Szilvia Szilágyi Ingrida Vaičiulytė

Lissajous figures are frequently studied and widely used objects in engineering and physics. Although these patterns are usually analysed using computer simulations or oscilloscopes, such tools may limit their educational value by covering the core physical processes that generate the curves. In order to address these problems, the design, kinematic validation, and prototyping of a dual-axis drawing robot were carried out on the LEGO Education SPIKE Prime platform. The hardware implementation centres on a LEGO-based dual Scotch yoke mechanism, which supports precise transformation of uniform circular motion into simple harmonic motion. This setup implements the superposition of two independent simple harmonic oscillations by simultaneously moving the paper tray along the x-axis and the pen along the y-axis. High-fidelity trajectories are achieved through a 40:1 worm gear reduction, which enables precise control of the parameter configuration. The phase shift can be manually set by adjustment levers. The robot’s geometry supports discrete amplitude settings of 8, 16, and 24 mm by adjusting the crankpin position. System control is managed by Python code that synchronises motor speeds and angular displacements according to frequency ratios. The research methodology used the Double Diamond design thinking framework, structuring development into four phases: identifying historical mechanical solutions, defining pedagogical and technical classroom requirements, iteratively developing the LEGO prototype, and testing the system through representative drawing experiments. Results show that the robot can reproduce a broad range of periodic Lissajous curves with high repeatability, and that its physical outputs show strong visual and mathematical correspondence to ideal trajectories simulated in the Desmos graphing calculator. The final prototype satisfies classroom constraints, providing a transparent, low-cost, modular STEAM tool that bridges the distance between abstract parametric equations and complex mechanical implementations.

Computers doi: 10.3390/computers15080528

Authors: Md Nahidur Rahaman Abdullah Al Mamun Md. Kamal Hossen Abdur Rouf Tumpa Rani Shaha Jungpil Shin Mohd Nizam Husen Abu Saleh Musa Miah

Rice leaf diseases significantly affect crop health and yield potential, creating a need for accurate and timely disease diagnosis. Many existing approaches still rely on single-stream feature extraction architectures, which may limit the ability to simultaneously capture global contextual information and fine-grained disease characteristics. Moreover, limited interpretability and decision-support capability hinder their practical deployment in real-world rice farming. To address these limitations, we employed a framework consists of two parallel feature extraction streams designed to capture different characteristics of disease patterns. The first stream uses a Swin Transformer to learn global contextual information and long-range spatial relationships across the leaf image. The second stream employs ConvNeXt to extract local texture features, including lesion details, spots, and color variations. By combining these complementary representations, the proposed framework effectively integrates global semantic information with local disease-specific features for improved classification performance. The extracted features from the two streams are fused through concatenation followed by an attention-based feature refinement module, enabling adaptive weighting of discriminative features. The refined representation is then used by a fully connected classifier for disease prediction. To enhance model interpretability, Grad-CAM visualization is incorporated to highlight disease-relevant regions and provide visual explanations for the model decisions. Furthermore, an LLM-based advisory module is integrated as a post-diagnosis decision-support component to provide contextualized disease management information and suggestions based on the predicted disease category. The generated suggestions are intended to support, rather than replace, expert agronomic recommendations and should be validated by agricultural professionals before practical application. The proposed framework was evaluated on two rice leaf disease datasets, achieving accuracies of 99.55% and 97.06% on Dataset-1 and Dataset-2, respectively, which are higher than those reported in previous studies. Additionally, five-fold cross-validation on Dataset-1 achieved an average accuracy of 98.91% ± 0.47, demonstrating the stability of the proposed approach. Cross-dataset evaluation using nine common disease classes across both datasets achieved 90.80% accuracy, indicating improved generalization across different data distributions. The proposed framework provides an accurate and explainable approach for rice leaf disease diagnosis in smart agriculture applications.

Computers doi: 10.3390/computers15080527

Authors: Haiyong Xie Lin Wang Yonghao Dai Yunqian Cheng

Fingerprint enhancement plays an important role in recovering identity-related ridge structures from degraded fingerprints. However, existing methods primarily focus on local texture restoration and may struggle to preserve ridge continuity and structural consistency under severe degradation conditions, including ridge fragmentation, diffusion blur, and partial information loss. In this paper, we observe that degraded fingerprints may retain incomplete ridge-flow information that can provide useful structural guidance for fingerprint reconstruction. Based on this observation, we propose a conditional generative adversarial network guided by a weak ridge-flow prior (WRP-cGAN) for degraded fingerprint enhancement. The proposed method treats the estimated ridge-flow information as a weak structural prior rather than an exact structural constraint and introduces prior-conditioned feature modulation to adaptively incorporate structural cues during reconstruction. The framework is jointly optimized using adversarial, image-space reconstruction, ridge-flow orientation-consistency, and gradient-consistency losses to improve ridge continuity, structural coherence, and local detail preservation. On the NIST SD301-derived test set, the proposed method increases the median NFIQ2 score from 9 to 44, improves the minutiae-restoration F1-score from 0.2507 to 0.5426, and increases the SourceAFIS Rank-1 identification rate from 39% to 86%. An additional qualitative evaluation on FVC2004 DB1 provides preliminary evidence of cross-dataset transferability without fine-tuning. These results suggest that weak ridge-flow priors provide useful structural guidance for degraded fingerprint reconstruction and improve recognition-oriented fingerprint quality under the degradation conditions considered in this study.

Computers doi: 10.3390/computers15080526

Authors: Panagiota Athanasiou Emmanuel Fokides

The growth of immersive virtual reality learning environments (IVRLEs) coincides with the emergence of numerous virtual educational agents that offer support to their users. Researchers suggested various terms to describe these agents; yet, this terminology fragmentation creates conceptual ambiguity, rendering them hard to compare. The purpose of this scoping review was to map out the terms used to describe educational agents in IVRLEs and look at how they have been conceptualized, the characteristics that have been used to differentiate them, and the educational roles and associated learning functions they have. In addition, the goal was to identify the requirements for the development of a relevant taxonomy. Arksey’s and O’Malley’s methodological framework and the PRISMA Extension for Scoping Reviews were applied. Out of the 1867 articles found in the Scopus, Eric, and LearnTechLib databases, published between 2015 and 2026, 129 met the eligibility criteria. The findings revealed substantial conceptual fragmentation, with similar educational entities being described using different terminology while identical terms were frequently applied to conceptually distinct systems. Characteristics such as intelligence, embodiment, interaction modality, adaptivity, affective and social characteristics, and context awareness distinguished these agents rather than terminology alone. Advances in artificial intelligence have further blurred the boundaries of traditional categories. Overall, this review provides a theoretical foundation for future research aimed at developing more coherent conceptual frameworks and standardized approaches to the design, classification, and evaluation of virtual educational agents.

Computers doi: 10.3390/computers15080525

Authors: Aruna Srinivasan Surabhi Narayan Aarnav Sandeep Deshmukh

Background: Image splicing forgery detection is one of the most challenging problems in the field of image forensics as it involves identifying and localizing suspicious regions that are created by integrating contents from one or more different sources. The accurate detection and classification of splicing forgery still remains a difficult task because of the existence of overlapping image regions, which makes it complex to differentiate authentic and tampered images. This overlap causes a lack of feature representation, making it difficult to precisely detect the tampering in images. Also, the decision-making process of the model is often a black box, which makes it challenging to interpret and understand the rationale behind its decisions. Methods: To address these challenges, a Convolutional Block Attention Module (CBAM)–U-Net with Cross-Connected Filters–Convolutional Neural Network (CCF-CNN) is proposed to achieve precise detection and localization of spliced regions. The CBAM enhances spatial and channel-wise attention, enabling accurate localization of forged regions. The dual-phase CCF-CNN is incorporated with cross-connected filters to differentiate between the authentic and tampered regions by extracting global and local features. Additionally, a surrogate heatmap mechanism is introduced using intermediate decoder features to generate patch-level visual explanations, enabling precise localization of the spliced regions, thereby improving the model’s transparency in decision-making. Results: The proposed CCF-CNN obtains a high accuracy of 99.84% on the CASIA 2.0 dataset and an accuracy of 95.63% on the MISD. Conclusions: Compared to traditional CNNs such as VGG, ResNet and attention-based interpretability algorithms, the proposed model obtains higher performance in terms of detection and interpretability.

Computers doi: 10.3390/computers15080524

Authors: Marwa O. Al Enany Mazen Hesham Elnahal Amira M. Gaber

Underprediction of high electrical demand can be more operationally consequential than an equally sized overprediction, yet standard point forecasting models are optimized primarily for average error. This case study evaluates a multivariate quantile Transformer as a safety-oriented next observation forecasting layer for a university laboratory. A timestamp-level audit identified 33,374 native measurements collected from 22 April to 12 December 2024 at a median interval of approximately 10 min. The final leakage-free pipeline uses only real observations, performs the chronological split before sequence generation, fits all scalers on training data only, and rejects windows containing gaps greater than 30 min. Persistence, fixed-order SARIMA, LSTM, GRU, CNN–LSTM, and an MSE-trained Transformer were evaluated on the same 6595-sample test period. GRU achieved the best deterministic accuracy (MAE 0.017744 kW; RMSE 0.023278 kW), whereas the proposed τ = 0.90 Transformer intentionally traded point accuracy (MAE 0.033830 ± 0.001133 kW) for asymmetric risk control. Across five independent runs, it achieved a pinball loss of 0.004453 ± 0.000069 kW, empirical coverage of 87.95 ± 1.01%, and a peak underprediction rate of 26.64 ± 5.32%, compared with 72.94–100% for the conventional benchmark outputs. Additional τ = 0.75 and τ = 0.95 experiments demonstrate the expected accuracy–safety trade-off. MAPE is not used as a primary metric because near-zero loads make percentage errors unstable. The results support the proposed model as a complementary upper quantile forecasting layer for this small, dynamic facility; they do not establish general performance at feeder or system scale.

Computers doi: 10.3390/computers15080523

Authors: Alaa O. Elhadi Saad M. Darwish Mahmoud A. Mahdi

Accurate crop-growth estimation from unmanned aerial vehicle (UAV) imagery is important for precision agriculture, but image-only models can struggle to represent seasonal context. This study evaluates whether combining UAV imagery with weather variables and elapsed time improves continuous rice growth-rate prediction in a single-site, publicly available rice-seedling dataset spanning multiple growing seasons. A Time-Aware Late Fusion (TALF) model is introduced in which a convolutional branch encodes image features, a multilayer perceptron encodes contextual features, and the two streams are merged only at the regression head. Relative humidity, wind speed, and elapsed time are used as contextual inputs after season-aware preprocessing. Evaluation is reported on a chronological multi-season split using internal ablations rather than external generalization claims. TALF achieved a mean absolute error (MAE) of 0.031, compared with 0.1455 for the optimized image-only baseline and 0.0890 for an early-fusion image-and-weather baseline. A secondary tolerance-based metric reached 94.9% under the reported threshold. The results indicate that weather and elapsed-time context improve prediction on this dataset and that separating image and tabular encoders until the final layers is a competitive multimodal learning design under the reported protocol.

Computers doi: 10.3390/computers15080522

Authors: Li Liu Chenhao Gu Kaifeng Geng

This study addresses a distributed flexible job shop scheduling problem with machine degradation and state-driven imperfect preventive maintenance, denoted as MD-SDIPM-DFJSP. A bi-objective model is developed to minimize makespan and total energy consumption by jointly optimizing job assignment, operation sequencing, machine selection, and processing speed. The model links processing speed with processing time, power consumption, and degradation increment, and uses a unified degradation bound to represent both degradation and reliability constraints. Preventive maintenance is treated as an imperfect recovery action and is generated according to machine states and idle-window conditions. To solve the problem, a degradation-aware multi-objective memetic algorithm (DMA) is proposed, incorporating four-layer encoding, state-driven decoding, hybrid initialization, knowledge-guided neighborhood search, and a speed-based adjustment operator. Numerical experiments show that Gurobi solved the small instance to optimality with a 0% optimality gap, and the resulting schedule satisfied the modeled production, maintenance, degradation, reliability, and energy accounting requirements. Across 84 combinations of instances and factory sizes, DMA achieved the highest HV in 72 cases and the lowest IGD in 62 cases. The Wilcoxon tests further confirmed its overall advantages over the three comparison algorithms in terms of both HV and IGD.

Computers doi: 10.3390/computers15080521

Authors: Amal Aabdaoui Najlae Idrissi

Wireless sensor networks (WSNs) are sophisticated monitoring systems that gather environmental data via wireless sensors. Numerous wireless sensors that have been placed to monitor and gather data on a particular environment make up these networks. Typically, a base station or central node wirelessly collects the data prior to analysis. As the battery in wireless sensors is both non-replaceable and non-rechargeable, it represents a key element. As a result, optimizing energy consumption in WSN has become a growing concern. One of the key challenges is consequently the creation of effective protocols for communication in WSNs. In this article, we provide a novel MSA (Mosquito Swarm Algorithm) technique for cluster formation and data routing. Simulations indicate that our proposed algorithm conserves the energy of the nodes and keeps them running for a greater number of survival rounds compared to LEACH (Low-Energy Adaptive Clustering Hierarchy) by a difference of 70.14%, PSO-R (Particle Swarm Optimization with routing) by a difference of 4.76%, and BA-R (Bat Algorithm with routing) by a difference of 2.52%. Our algorithm provides the highest throughput, surpassing LEACH by approximately 6.38%, PSO-R by almost 1%, and BA-R by 12.67%. Simulations indicate that our algorithm is highly effective at extending network longevity and increasing throughput, making it the preferred option to lower energy consumption in WSNs.

Computers doi: 10.3390/computers15080520

Authors: Irum Naz Sodhar Dil Nawaz Hakro Abdul Hafeez Buller Umair Ramzan Sheikh Suad Mohammed Al Qassabi Osama Al Rahbi Akhtar Hussain Mohammed Izaan Kari

Sindhi language models struggle with Romanized Sindhi as there is variation in the spellings of everyday words, and short prompts do not explicitly state the task or target language. In this study, the remaining source record of a small DeepSeek-R1 probe, which was based on seven canonical prompts and fourteen alleged A/B records, was audited. The screenshot captions were compared with the visible inputs and outputs, and duplication of images was checked. It was only relabeled when there was a single unambiguous canonical prompt that matched the visible input to a record. Task choice and exact target-string occurrence were then coded separately. Six records were auditable, three of which were recovered by relabeling; eight records were not auditable due to being duplicated, contaminated with a supplied answer, associated with a different phrase, or missing the user input or lost after label repair. Four of the six audited records adhered to the experimenter’s task, one provided a reasonably good English translation when the experimenter did not include the target language, and one resulted in an incompatible cross-linguistic reading. Five had the target (English/Sindhi) string and one did not. The screenshots do not demonstrate clear “standard” and “chain of thought” conditions: all audited B-labeled inputs are in the form of a simple prompt and there are no paired wrapper texts remaining. These are descriptive counts, not accuracy estimates and do not provide proof of causal benefit of prompt explicitness. The audit demonstrates the necessity of full prompt–response documentation, independently verified linguistic references, reruns, information-matched prompts, and preregistered evaluation plans in studies with low resources.

Computers doi: 10.3390/computers15080519

Authors: Mohamed Eassa Nagwa Yaseen Hegazy Hussam Elbehiery

Detection of brain tumors through MRI scans is a difficult and crucial process since medical imaging is complex and there are not enough experts to analyze these images in many countries. The timely detection of diseases such as gliomas, meningiomas, and pituitary tumors is critical because delayed detection may have adverse impacts on the patient’s condition and result in incorrect treatment. These difficulties are why automated deep learning models were introduced to help diagnose brain tumors. In this research, we introduce a multi-stage sequential pipeline for brain tumor classification, localization, explainability, and report generation in a radiologist-style structured format based on MRI imaging data, with each stage trained and evaluated independently on its respective dataset. Specifically, an Xception-based classifier was used to classify MRI images into one of four classes, achieving 98.96% accuracy and an F1-score of 0.9876. To enhance interpretability, the Grad-CAM technique was used to visualize image patches the model used during prediction. If the tumor was present, U-Net was used to perform localization, giving a Dice score of 0.7835 and an IoU of 0.6919 with the use of T1-weighted MRI images. Finally, the obtained features were passed as input to a QLoRA fine-tuned language model that generates structured radiology reports, including such components as Technique, Findings, and Impression.The proposed framework was evaluated on a hold-out subset of publicly available T1-weighted MRI data, achieving LLM-as-a-Judge scores of 4.92/5 for accuracy and 4.95/5 for fluency. These results suggest that the proposed approach is promising and provide an initial proof of concept. However, additional validation on independent multi-center datasets and multimodal MRI data is still needed before considering its use in clinical practice.

Computers doi: 10.3390/computers15080518

Authors: Franklin Parrales-Bravo Joan Gracia-Chinga Janio Jadán-Guerrero Leonel Vasquez-Cevallos Lorenzo Cevallos-Torres Leili Lopezdominguez-Rivas

This scoping review synthesizes 35 studies on sensor technologies and artificial intelligence for fruit detection, classification, and quality assessment, contrasting Latin American and international research traditions. The analysis suggests notable differences in technological approaches between the included Latin American and international studies: Latin American research tends to emphasize in developing accessible, practical solutions using classical computer vision and low-cost hardware, while international studies more frequently employ through deep learning architectures, multi-modal sensing, and complete robotic automation systems. Across the included studies, relatively limited attention was given to AI-assisted decision support for agricultural practitioners, insufficient consideration of inclusivity, and the scarce integration of environmental sustainability into intelligent sensing system design. The review identifies that only 8 of 35 studies originate from Latin America, suggesting an uneven geographical distribution of the available evidence. The included studies generally reported high accuracy values, yet these findings must be interpreted with caution given the reliance on curated datasets that may not represent real-world variability. The reviewed evidence suggests that future research may benefit not from one approach dominating the other, but from a thoughtful integration of complementary strategies, including knowledge transfer, edge computing democratization, and human-centered design. Overall, this review suggests that the ultimate goal extends beyond accuracy metrics to the transformation of agricultural practices that enhance food security, economic development, and environmental sustainability across the global agricultural landscape. It is important to note that this work does not propose or validate a new fruit detection algorithm but rather synthesizes and critically evaluates existing scientific evidence regarding sensor technologies and artificial intelligence applied to fruit detection and quality assessment.

Computers doi: 10.3390/computers15080517

Authors: Yuetong Chen Yumei Wang Yufu Ning Fengming Liu Mingrui Zhou

Public opinion event response requires not only timely decisions but also transparent and trustworthy collaboration among multiple stakeholders. To address delayed responses, fragmented collaboration, and information opacity, this study first proposes a collaborative-decision model based on blockchain for public opinion event response and then develops a blockchain-driven adaptive consensus method to improve consensus efficiency and decision quality. In the proposed model, public opinion information is mined to identify the attribute categories and weights of response alternatives, while collaborative-decision quality is evaluated by integrating decision reliability, opinion convergence, and individual comprehensive weights derived from social network influence. On this basis, smart contracts are designed to support transparent, traceable, and automated consensus processes. The adaptive consensus method dynamically terminates the consensus process by considering public opinion crisis levels and individual consensus differentiation. A utility-maximizing feedback mechanism is further introduced to improve consensus quality, and smart contracts are used to detect the adjustment willingness of inconsistent individuals and implement an elastic incentive mechanism. Case analysis and simulation experiments verify the effectiveness and robustness of the proposed model and method, showing their potential to support trustworthy collaborative decision-making in public opinion event response under uncertain and time-sensitive conditions.

Computers doi: 10.3390/computers15080516

Authors: Stefan Reitmann Tsvetomila Mihaylova Dionysios Kritharoulas Eelis Peltola Elin A. Topp Ville Kyrki

In level-3 automated driving, control shifts between the system and the human driver, making timely and effective communication crucial in conflict situations. To explore how different communication modalities affect the driver’s understanding, trust, and performance in such situations, we conducted two complementary studies: a focus group and an interactive user study. The focus group revealed a preference for multimodal, tailored feedback, with visual information most frequently favored. The interactive user study tested these findings in practice by asking participants to confirm the system suggestion or take over. The results from the interactive study showed the importance of a pre-explanation signal that draws attention to the situation, and a clear visual marking of the proposed resolution. While most participants stated a preference for speech-augmented feedback, spoken and written explanations were frequently overlooked or reported as distracting during active conflict scenarios, suggesting a gap between stated preference and in-task utility. These results indicate that in urgent conflict situations, concise visual cues are more effective than detailed verbal explanations, offering guidance for the design of future level-3 vehicle interfaces.

Computers doi: 10.3390/computers15080515

Authors: Marcela Guerrero-Calvache Giovanni Hernández María Clara Gómez-Álvarez

Measuring productivity in software development teams is a key process for evaluating their performance in agile environments, which are characterized by value creation, continuous adaptation, and incremental improvement. However, in the field of software engineering, although there are approaches focused on individual metrics, there remains a gap in the development of multifactorial models that integrate the various dimensions influencing team productivity. In response to this issue, this study proposes a multifactorial conceptual model designed to support the process of measuring productivity in agile teams. The development of the model was based on the measurement protocol proposed by Fenton and Bieman and included two main stages: design and validation. During the design phase, productivity factors were identified, and entities, properties, and empirical relationships were defined; these were represented using a Unified Modeling Language (UML) class diagram. The model was validated through expert judgment and an exploratory empirical application in higher education settings. The results demonstrate a high level of acceptance by experts, as well as the model’s viability for application in real-world scenarios, enabling the operationalization of the productivity construct in Scrum teams. In conclusion, the proposed model constitutes a significant advance in measuring productivity in agile teams by integrating multiple dimensions of performance, offering a structured, validated, and applicable framework that overcomes the limitations of traditional approaches and contributes to both the academic realm and professional practice.

Computers doi: 10.3390/computers15080514

Authors: Abdalbasit Qadir Bryar A. Hassan Hozan Khalid

Deploying accurate face recognition models on resource-constrained devices remains a significant challenge. Ghost modules mitigate feature-map redundancy by synthesizing half of each feature map through inexpensive linear transformations and form the basis of the lightweight GhostFaceNet and GhostFaceNet++ families. Nevertheless, the inexpensive generator is a strictly local 3 × 3 depthwise convolution: the synthesized features remain near-duplicates of the intrinsic ones, and deeper stages lack the global receptive field required for cross-pose and cross-age matching. This paper introduces FD-GhostFaceNet, built upon the proposed Frequency-Decoupled Ghost (FD-Ghost) module, which replaces the local generator with a complementary-band global counterpart. A learnable, input-conditioned partition of the two-dimensional Discrete Cosine Transform spectrum decomposes the intrinsic features into a global low-frequency band that encodes pose- and age-stable identity structure and a locally refined high-frequency detail band. Nyquist-consistent anti-aliased downsampling, which band-limits each feature map before subsampling so that the decoupled high-frequency band survives decimation, completes the frequency-decoupled design. Applied to both published trunks, FD-GhostFaceNet-V1-2 and FD-GhostFaceNet-V2-2 require 4.60 M and 7.39 M parameters, 83.1 and 97.1 MFLOPs, and 9.19 and 14.77 MB of storage, respectively, and thus fall within the sub-100 MFLOPs category of lightweight face recognition models. Across six standard benchmarks, both variants surpass their corresponding GhostFaceNet baselines in 23 of 24 trunk–benchmark comparisons and are competitive with or superior to the GhostFaceNet++ variants, with the largest gains on the pose-sensitive evaluations. When trained with ArcFace on UMDFaces, FD-GhostFaceNet raises the best reported CP-LFW accuracy from 84.65% to 86.12% and CFP-FP from 87.60% to 89.56%; when trained on CASIA-WebFace, it improves the best reported CFP-FP from 90.10% to 90.94%. These results confirm that frequency decoupling strikes a favorable balance between compactness and accuracy, making both models strong candidates for deployment on resource-constrained devices.

Computers doi: 10.3390/computers15080513

Authors: Anibal Flores Ruso Morales-Gonzales Jose Guzman-Valdivia Saul Huaquipaco Carlos Silva-Delgado Hugo Tito-Chura Mario Gauna-Chino Honorato Ccalli-Pacco

Ring-neck disease in avocado cultivation is a critical issue due to its significant impact on production, including economic losses caused by premature fruit drop, postharvest rejection, and export limitations, among other factors. To enable the automated detection of ring-neck symptoms, this study proposes a computer vision-based framework. As an initial step, a dataset of 622 Fuerte avocado images was established and annotated, consisting of 361 healthy samples and 261 samples affected by ring-neck. During the experimental stage, convolutional neural network (CNN) models with different architectures, comprising four and five convolutional layers, were developed and systematically evaluated. In addition, a region-of-interest (ROI)-based strategy was employed to focus the analysis on the fruit peduncle, thereby enhancing the detection of ring-neck symptoms. The performance of the proposed model was benchmarked against widely adopted deep learning architectures, namely VGG-16, VGG-19, ResNet-18, MobileNetV3, and GhostNet. The experimental evaluation demonstrated the superiority of the proposed approach, achieving an F1-score of 0.9610, compared with 0.5667 for VGG-16, 0.7164 for VGG-19, 0.0952 for ResNet-18, 0.6176 for MobileNetV3, and 0.7733 for GhostNet. Furthermore, statistical significance was assessed using the McNemar test, which confirmed that the observed performance improvements over the benchmark models were statistically significant, thereby supporting the robustness and reliability of the proposed approach.

Computers doi: 10.3390/computers15080512

Authors: Faris Kateb Owais Khan Fazal Qudus Khan

With the rapid expansion of the Internet of Medical Things (IoMT), the risks of cybersecurity have increased exponentially in healthcare settings, exposing patients’ safety. Three fundamental issues that existing intrusion detection systems (IDS) are challenged by are: (1) limited cross-domain generalization, (2) high computation requirements not suitable for edge deployment, and (3) absence of systematic comparison between classical machine learning (ML) and deep learning (DL) approaches on IoMT-specific data. In this paper, we propose a multi-dataset evaluation framework that covers six models (Random Forest, XGBoost, DNN, CNN, LSTM, and CNN-LSTM) across three different datasets: WUSTL-EHMS-2020, Edge-IIoTset, and UNSW-NB15. We show that there is a scale-dependent pattern: classical ensemble methods work best when the data is small (F1 = 0.914 ± 0.013 on WUSTL-EHMS-2020); the proposed hybrid CNN–Ensemble framework performs best when the data is large (F1 = 0.968 ± 0.002 on UNSW-NB15 with 62.8% fewer features). The proposed framework achieves a total model size of 2.11 MB and an inference latency of 111.6 ms, with seven out of the top 15 discriminative features being patient vital signs, giving the first quantitative evidence that physiological data systematically contributes to IoMT attack detection, which is demonstrated through an explainability analysis using the SHAP approach. Cross-dataset generalization experiments across six transfer scenarios expose fundamental limitations in domain transfer, establishing an important baseline for future research.

Computers doi: 10.3390/computers15080511

Authors: Leandros A. Maglaras Christina Kyritsi Helge Janicke Konstantinos Karantzalos

Mobile-phone overuse and attention fragmentation have become pressing societal and public-health concerns. Cyberpsychology research highlights addictive engagement loops driven by intermittent rewards, persuasive design, and habit formation. In this article we synthesize current evidence on mobile-phone addiction and propose “Digital White Spaces” (DWSs), a socio-technical framework that combines privacy-preserving monitoring, AI-driven detection of addictive loops, device-mode interventions, and physical signal-limited zones to restore user autonomy.

Computers doi: 10.3390/computers15080510

Authors: Robert Campbell

Agent identity governance governs an agent’s identity, credentials, and lifecycle, but assumes the composition it was approved with is the composition it runs with. That stability assumption is unenforced: no lifecycle mechanism evaluates an agent’s evolving composition against its approved baseline. An agent’s effective composition—tools, data sources, delegated authorities, child agents—is a runtime supply chain of capability, and it drifts. We introduce composition drift, the departure of effective composition from the terms of approval, and isolate its sharpest form, compositional drift: individually approved changes accumulating into a capability none authorized alone. We formalize this with a two-stage operator: a component diff detects that the composition changed; a capability-closure stage detects when it authorized something new. The contribution is a temporal governance model linking emergent capability to reauthorization and inventory reconciliation. Drift produces shadow infrastructure: resources provisioned outside any inventory through benign, individually approved pathways. We propose composition attestation, a runtime control complementary to identity governance, and evaluate it in a pre-registered study. Across five thousand trajectories the detector separates compositional drift from authorized growth where a static analyzer cannot; an ablation isolates the primitives that cause it; and, in controlled live agent runs across three models (9B to 70B parameters, Llama and Qwen lineages), agents given only benign tasks provision unauthorized shadow infrastructure in 59% to 99% of drift-conducive trials against 0% to 19% of matched controls.

Computers doi: 10.3390/computers15080509

Authors: Huixing Fang

We present a compositional formal verification framework that establishes floating-point correctness and control safety for anti-lock braking system (ABS) neural controllers, using Rocq and Flocq as a single verification toolchain. We decompose the verification into three independent components. First, a Flocq proof bounds each binary32 operation to machine epsilon. Second, a CoqInterval-verified barrier certificate enforces eight pointwise conditions across four road surfaces, keeping the slip ratio within [0.03, 0.40]. Third, an ODE forward invariance lemma lifts these pointwise conditions to a temporal guarantee via the suprema axiom and ε-δ continuity arguments. The compositional theorem combines all three layers into an end-to-end safety proof. Our Rocq development (866 lines, 7 modules) compiles successfully: the core floating-point lemmas are Qed, with one lemma remaining Admitted. Numerical simulation across four road surfaces (after 0.3 s settling, v≥1.5 m/s) shows that the neural network controller achieves 100% closed-loop safety on every surface, including ice, outperforming the classical finite state machine controller (96.0–100%). The framework demonstrates that single-toolchain compositional verification of floating-point neural controllers is feasible, directly supporting ISO 26262 certification for safety-critical automotive systems. The verified barrier windows and open-source Rocq artifacts provide a reusable foundation for future neural ABS verification work.

Computers doi: 10.3390/computers15080508

Authors: Mariappan Kadarkarainadar Marichelvam Mariappan Geetha

A hybrid algorithm combining two metaheuristics is proposed to solve the flow-shop scheduling problem, aiming to minimise the makespan (Cmax). This approach accounts for random machine failures and limited buffer capacity between machines. Since flow-shop scheduling problems are NP-hard, the metaheuristics could be used to solve them effectively. Researchers proved that the hybridisation of metaheuristics would improve the solution quality. Therefore, this study hybridises the recently developed Pigeon-Inspired Optimisation Algorithm (PIOA) with the artificial bee colony (ABC) algorithm. The initial solutions are generated using a dynamic generation technique that relies on a set of constructive heuristics. The optimal solutions from the PIOA serve as input for the ABC algorithm. Various local search and variable neighbourhood search methods are also included to enhance solution quality. Extensive computational experiments, which focus on industrial scheduling scenarios and benchmark problem instances, are conducted to test the performance of the hybrid algorithm. Statistical analysis shows that the proposed algorithm outperforms other algorithms found in the existing literature.

Computers doi: 10.3390/computers15080506

Authors: Angelower Santana-Velásquez Maria Bernarda Salazar-Sánchez

Explainable Artificial Intelligence (XAI) has emerged as a critical enabler for the adoption of machine learning models in high-stakes domains such as healthcare. While significant progress has been made in XAI for computer vision and natural language processing, tabular data—the predominant format of electronic health records—presents unique challenges and opportunities. This systematic review provides a comprehensive analysis of XAI methods specifically applied to tabular healthcare data for classification tasks. We examine 21 primary studies published between 2020 and the first half of 2026, covering three complementary perspectives: (1) intrinsically interpretable models, (2) post-hoc methods including LIME, SHAP, and their variants, and (3) evaluation frameworks that assess both model-centered fidelity and human-centered clinical alignment. Our analysis reveals that SHAP remains the dominant post-hoc method, achieving strong model fidelity but showing inconsistent alignment with clinical expert reasoning. Key findings include the significant impact of class imbalance on explanation consistency, the importance of clinician-centered evaluation, and the emergence of hybrid approaches integrating XAI with generative AI and transfer learning. We identify critical gaps, including limited adoption of XAI in AutoML pipelines, lack of standardized evaluation metrics, and predominance of single-institution validation studies.

Computers doi: 10.3390/computers15080507

Authors: Dalma Lilla Dominek Vanessza Kapusi Szabolcs Ceglédi Zoltán Szűts

The rapid adoption of generative artificial intelligence (AI) in higher education has raised important questions about its impact on students’ cognitive processes. While AI can support learning and problem-solving, concerns have emerged regarding its influence on independent thinking. This study examines whether students perceive AI primarily as a cognitive complement or as a replacement for their own thinking, and how these perceptions relate to cognitive outcomes. Data were collected from N = 93 university students at a Hungarian university; analyses were conducted on the subsample of n = 73 AI users. Exploratory factor analysis identified three cognitive dimensions, and correlation and regression analyses examined relationships among AI use patterns, perceived AI role, and cognitive outcomes. The results indicate that perceiving AI as a complement to one’s own thinking is positively associated with cognitive independence and represents its strongest predictor. In contrast, neither the frequency nor the specific purposes of AI use significantly predicted cognitive independence. An exploratory analysis further revealed a positive association between AI use for coding and problem-solving and perceived cognitive augmentation. Nonetheless, given the exploratory nature of the study and the single-institution sample, these insights should be regarded as preliminary. Additionally, the cognitive augmentation framework (H3) yielded only partial empirical support, characterized by a single significant predictor embedded within an overall non-significant model. These findings suggest that the cognitive consequences of AI depend less on usage intensity and more on how students conceptualize AI’s role in their thinking processes.

Computers doi: 10.3390/computers15080505

Authors: Hadeel Amjed Saeed Sufyan T. Faraj Al-Janabi Esam Taha Yassen Omar A. Aldhaibani

A reinforcement learning hyper-heuristic framework for multi-objective scientific workflow scheduling selects among five meta-heuristic optimisers and tunes their control parameters under a Nash Social Welfare reward over makespan, cost, security, and resource utilisation. In its base form it treats security as a static virtual-machine attribute and admits all candidates unconditionally. This paper contributes the mechanism design required to integrate two security-realism layers into the scheduling loop without redesigning the reward: a five-stage zero-trust admission pipeline, a bounded non-stationary per-machine dynamic trust signal, a state-vector augmentation that exposes trust to the agent, and a coupling that attenuates the effective security level seen by the security utility. The two layers act at distinct timescales: admission is a provisioning-time gate on a machine’s structural compliance, whereas the trust signal evolves per decision epoch for the machines already admitted, so static admission and dynamic trust coexist by construction. We evaluate three hyper-heuristic agents on 20 Pegasus workflow instances under both a trust suite and a trust-free baseline. The base framework establishes a sharp separation between the hyper-heuristic and direct task-to-machine RL families; we treat this as an inherited property and ask a different question: can the two security-realism layers be integrated without disturbing it? Across 20 Pegasus instances and three HH-RL agents, the family-level separation is preserved. Under a reproducible evaluation protocol—five independently seeded repeats of the full paired comparison, 100 greedy inference episodes per (agent, workflow, suite) cell, with per-workflow deltas averaged across repeats before testing—the trust extension imposes a small, heterogeneous absorption cost: the median per-workflow shift in Nash reward is −0.24, −0.24, and −0.05 for PDQN, DQNHH, and QLHH respectively, an order of magnitude below the absolute reward levels. The shift is statistically significant for DQNHH (two-sided Wilcoxon p = 0.0014, rank-biserial r = −0.77), marginal for PDQN (p = 0.058), and absent for QLHH (p = 0.18). The security utility stays above 0.91 on every instance, and the family-level scaling robustness is preserved intact. The contribution is therefore a drop-in mechanism whose cost is bounded and small relative to the between-family separation—with a robust workflow-level heterogeneity: the parameterised agent converts the trust signal into consistent gains on the largest DAGs (mean +1.28 on Sipht_1000 and Inspiral_1000 across the five repeats) while paying a small cost on typical instances.

Computers doi: 10.3390/computers15080504

Authors: Ali Fenjan Mohammed Almulla Jalil Md. Desa

Android malware detection systems commonly emphasize predictive accuracy while paying less attention to feature-acquisition cost, deployment efficiency, and early decision making. This paper presents a staged model-input budget framework for cost-aware Android malware screening and evaluates classification performance under progressively expanded static feature representations. The proposed framework uses a lightweight Behaviour Score as an early-warning model input derived from multiple static behavioural indicators, including permission risk, encryption evidence, network activity, and suspicious keyword evidence. Rather than treating the score as a cost-free feature, the framework distinguishes between the derived model input and the underlying static indicators required to construct it. Uncertain samples are progressively escalated from the early-warning stage to richer feature budgets using a confidence-based decision rule, while confident samples can be resolved before full-feature analysis. The experimental evaluation reports hyperparameter-tuned model performance, empirical inference-time profiling, confidence-based escalation behaviour, Matthews correlation coefficient, false positive rate analysis, low false-positive-rate operating points, cross-validation, statistical testing, and external proxy-budget validation using the Drebin benchmark. On the main Android application dataset, the Behaviour-Score stage achieved an F1-score of 0.8750. When low-cost static indicators were added, the framework achieved an F1-score of 0.9654 and a Matthews correlation coefficient of 0.9267. The full feature set achieved the highest F1-score of 0.9878 and Matthews correlation coefficient of 0.9741. The confidence-based escalation experiment showed that, at a predefined 0.95 confidence operating point, 90.32% of samples were resolved before full-feature analysis, reducing the average number of classifier model inputs used from 9 to 3.50 while maintaining an F1-score of 0.9785. These findings indicate that the proposed framework provides an incremental model-input budget approach for deployment-oriented Android malware screening, while preserving full analysis for uncertain samples.

Computers doi: 10.3390/computers15080503

Authors: Antonio Carlos Bento Elsa Yolanda Torres-Torres Sérgio Camacho-León Carlos Vázquez-Hurtado Bárbara Martínez-Mijares Fernanda Santillán-Dantés Marcelo Guillé-Martínez Ximena Villarreal-Solórzano Brian Roberto Gómez-Martínez

Project-Based Learning (PBL) has become an important pedagogical strategy for developing technical and professional competencies in engineering education through authentic, multidisciplinary experiences. This paper presents a PBL case study conducted in an undergraduate Computer Engineering course in which students designed and implemented Smart Fisio Borregos, an Internet of Things (IoT) and Artificial Intelligence (AI) system designed as both a learning vehicle for students and a prototype tool intended to support rehabilitation training through real-time exercise guidance and monitoring; the educational effectiveness of the rehabilitation-support function has not yet been formally validated. The project followed a prototype-driven methodology that integrated low-cost sensors, embedded systems, cloud databases, computer vision, and AI services into a unified platform. The resulting prototype incorporated equipment occupancy monitoring, environmental control, RFID-based access management, a web dashboard for data visualization, and a computer-vision module based on MediaPipe Pose Landmarker for exercise analysis and feedback. The project provided students with opportunities to apply knowledge from programming, embedded systems, databases, networking, and AI while developing collaboration, problem-solving, and project-management skills. The paper describes the pedagogical framework, system architecture, implementation process, and project outcomes, illustrating how multidisciplinary engineering projects can be used to create authentic learning experiences connected to real-world challenges. The proposed approach offers a replicable model for integrating IoT and AI technologies into engineering curricula while contributing to educational innovation related to health and well-being. The study aligns with Sustainable Development Goal 3 (Good Health and Well-Being) and Sustainable Development Goal 9 (Industry, Innovation, and Infrastructure).

Computers doi: 10.3390/computers15080502

Authors: Didar Hossain Mohiuddin Mehedi Khandakar Rabbi Ahmed Md Rafiul Mahmud Mainul Islam Khan Sakib Salam Jamee

Customer churn is a critical challenge for subscription-based digital platforms, as customer activity patterns change dynamically over time. Traditional churn modeling methods fail to account for sequential dependencies across customer interaction histories. This paper presents Attn-ChurnNet, a novel attention-based Transformer architecture that effectively predicts customer churn by modeling sequential customer contact histories. The proposed methodology leverages multi-head self-attention with sinusoidal positional encodings and Pre-LN residual connections to highlight key interaction sequences and interpret the temporal dynamics of customer activity in subscription platforms. Experiments are conducted on the large-scale WSDM–KKBox Customer Churn Prediction dataset using a temporal train/validate/test split, incorporating transaction records, usage logs, and customer demographic information. Comprehensive comparison against established baselines—Logistic Regression (LR), Random Forest (RF), XGBoost, and Gated Recurrent Unit (GRU)—demonstrates that Attn-ChurnNet achieves a macro-averaged classification accuracy of 97.03% (±0.41%), precision of 95% (±0.5%), recall of 96% (±0.6%), F1-score of 95.50% (±0.5%), AUC of 0.98 (±0.004), Average Precision of 0.963, and log-loss of 0.15 (±0.007) under five-fold stratified cross-validation, outperforming all competing approaches with statistical significance (p<0.01, McNemar’s test). A comprehensive two-part ablation study (22 variants), calibration analysis (ECE = 0.031; Ts* = 1.08), attention entropy analysis with Jensen–Shannon divergence and two-sample t-test (t=18.4, p<0.001), Integrated Gradients attribution, conformal prediction (91.4% coverage, 88% singleton efficiency), precision–recall analysis, and computational complexity evaluation further validate the model’s design and production readiness.

Computers doi: 10.3390/computers15080501

Authors: Md Nuruzzaman Pranto Md Deluar Hossen Mamunur R. Raja Md Sharfuddin Balayet Hossain Khandakar Rabbi Ahmed

Due to the explosive growth in cloud-native infrastructures, the attack surface has dramatically increased in modern enterprise identity systems, where privilege escalation has become a major security risk. Conventional rule-based intrusion detection systems fall short in identifying multi-hop privilege inheritance paths and lateral movements over heterogeneous and dynamic identity graphs. This study introduces PEGraphSec-Net, a graph-theoretical framework for detecting privilege-escalation-relevant identity behavior, modeling cloud identity interactions as dynamic heterogeneous graphs of users, services, roles, tokens, and workloads. The core contribution of this framework is a graph-based detection pipeline—an Identity Relationship Graph Constructor, a Privilege-Escalation Path Encoder, and a Temporal Graph Attention Detection layer—evaluated on privilege-escalation-relevant attack categories using a documented proxy identity-graph construction derived from the UNSW-NB15 network-traffic benchmark, and benchmarked against six non-graph tabular classifiers (CNN, LightGBM, XGBoost, Random Forest, SVM, and MLP) trained under identical preprocessing; this pipeline achieves 98.78% accuracy, a weighted F1-score of 0.98692 (macro F1-score of 0.91828), and an AUC of 1.000 on the held-out test partition. PEGraphSec-Net is further benchmarked against three graph neural network baselines (GCN, GAT, and GraphSAGE) trained on the identical identity-graph topology and node attributes; all three substantially underperform PEGraphSec-Net (best case, GraphSAGE: 63.66% accuracy, 0.239 macro F1-score), indicating that a large share of PEGraphSec-Net’s performance derives from its explicit privilege-path encoding and temporal attention mechanisms rather than from the graph topology alone. An Adaptive Containment and Isolation Engine and a Mitigation Policy Reinforcement Optimizer are further proposed as risk-scoring and reward-driven policy-learning components, whose contribution is validated through module-wise ablation on classification performance; live containment action and reinforcement-learning-specific evaluation are left for future validation. The term “privilege escalation” is used throughout to denote the evaluated proxy attack categories (Exploits, Backdoor/Backdoors, and Reconnaissance) under a documented, decade-old (2015) network-intrusion benchmark, rather than production cloud-native IAM behavior, for which native-dataset validation remains an open direction. SHAP-based interpretability analysis links the model’s top-ranked traffic-level features back to the identity-graph risk, role, and trust-transition attributes they populate, evidencing that the learned representation captures semantically meaningful identity-behavior patterns within this proxy setting.

Computers doi: 10.3390/computers15080500

Authors: Rafeek Mamdouh Ahmed Hagag Ramadan Babers

Demand for high-quality interactive and real-time rendering remains challenging, as it requires balancing image realism with computational resources. Static parameter tuning of traditional approaches cannot provide adaptive rendering according to the varying complexity of dynamic scenes. This limitation arises from two main deficiencies in existing rendering pipelines: reactive methods that only enhance images after rendering without optimizing the renderer itself, and proactive methods that still rely on manual parameter calibration for each scene. These shortcomings are solved by this paper with an innovative optimization method that is a combination of a genetic algorithm (GA) and artificial neural networks (ANNs). This method offers a closed-loop system that is not found in any other static pipeline. Specifically, in our approach, ANNs will be used to predict the renderer’s initial parameter values from scene descriptor data, such as the number of polygons, lighting, and materials. After predicting the parameters, GA will optimize them based on the fitness value, which is determined by maximizing one objective (perceptual quality, defined by the SSIM measure) and minimizing another (rendering time). Our approach can be easily implemented within standard pipeline frameworks (Autodesk Maya Arnold).

Computers doi: 10.3390/computers15080499

Authors: Md Tahmid Rashid Md Jawad Siddique

UAV-assisted cyber–physical–social systems increasingly use visual analytics for disaster response, infrastructure monitoring, traffic management, and safety-oriented situational awareness. Raw aerial imagery can expose faces, license plates, private-property context, location traces, and sensitive human activity, while cloud-based vision–language processing can increase latency, bandwidth use, and governance risk. This article presents PrivEdge-VLM, a risk-adaptive privacy-preserving edge vision–language analytics framework for UAV-assisted cyber–physical–social systems. PrivEdge-VLM integrates object detection, privacy-sensitive region detection, mission-aware semantic tokenization, local inference, sanitized split inference, federated LoRA adaptation with differential privacy and simulated masked aggregation, and deterministic operator guardrails. Across public UAV imagery, public privacy benchmarks, and non-human staged object protocol examples, the evaluation compares raw-cloud, downsampling, static-blur, local-VLM, raw-embedding split, federated static-privacy, no-DP, and full configurations. The full configuration preserves mission utility while reducing residual visual PII, plate-OCR leakage, membership-inference risk, embedding inversion, bandwidth, and unsafe identity-oriented responses under the stated attack protocols. The edge hardware deployment separates privacy-front-end throughput from event-level VLM query latency. The results indicate that UAV VLM privacy is a CPS co-design problem involving semantic transformation, attack-based leakage evaluation, resource-aware deployment, and operator oversight.

Computers doi: 10.3390/computers15080498

Authors: Xuefei Huang Zihan Zhang Kyriakos M. Deliparaschos

Incremental two-dimensional (2D) Delaunay triangulation is a fundamental operation in point-cloud processing and surface reconstruction, yet its irregular memory access and dynamically evolving topology make efficient hardware implementation difficult. Conventional point-location strategies validate many candidate triangles, limiting efficiency on resource-constrained embedded platforms. This paper presents a unified System-on-Chip Field-Programmable Gate Array (SoC-FPGA) framework that instantiates two design points: an acceleration-oriented variant that lowers latency for small-to-medium inputs and a scalability-oriented variant that extends input capacity. Both share a common backbone: a Doubly Connected Edge List (DCEL) topology, a structure-of-arrays memory layout with heterogeneous on-chip binding, a Hilbert-ordered insertion sequence, and a grid-assisted scheme that turns global point location into a bounded local search. Implemented on the AMD Kria KR260, the deployed PS–PL system executes the Delaunay triangulation in the programmable logic, while the ARM Cortex-A53 processor manages accelerator control, data transfer, and end-to-end timing. For a 6000-point input, the acceleration-oriented variant requires 482.25 ms in RTL co-simulation, corresponding to a 32.7× speedup over the same-flow Vitis C-simulation reference on an Intel workstation. Its measured PS–PL end-to-end runtime is 315.919±0.042 ms, achieving a 1.96× speedup over a CPU-only Cortex-A53 implementation on the same KR260 platform and a separately reported 50.0× speedup over the workstation Vitis C-simulation reference. The scalability-oriented variant completes the same input in 868.51 ms and reduces the point-location fallback rate from 12.6% to 2.9%. On a 16,000-point set that exceeds the acceleration-oriented variant’s on-chip budget, it finishes in 4360.6 ms in RTL co-simulation, corresponding to a 19.0× speedup over its same-flow workstation Vitis C-simulation reference. The framework thus provides explicit, tunable trade-offs between latency and scalability within a single architectural template, offering a practical solution for embedded geometric preprocessing.

Computers doi: 10.3390/computers15080497

Authors: Stelios Xinogalos

Programming education has attracted the interest of instructors and researchers for several decades [...]

Computers doi: 10.3390/computers15080496

Authors: Nicolás Baier Quezada Vanessa Uribe Hernández Haydée Barrientos Toledo Cristina Vargas Bustamante Martin Arrigo Figueroa Aaron Mancilla Leiva Felipe Brana Peña Fernanda López-Moncada

Annotated dataset preparation remains a critical bottleneck in machine learning (ML) pipelines. Existing annotation tools—cloud-hosted services, self-hosted web applications, and lightweight desktop editors—each cover part of this workflow, but few combine broad annotation-type support with offline operation, integrated training, and serverless collaboration. We present Annotix, an open-source, cross-platform desktop application that brings the complete ML data preparation workflow—annotation, training, model-assisted labeling, and review—into a single privacy-preserving environment. It is implemented with a Rust/Tauri 2 backend and a React 19 frontend, and runs locally without server infrastructure or cloud services. Annotix was evaluated through three complementary studies. First, a comparative feature analysis positioned the platform against established annotation tools. Second, a controlled efficiency experiment compared Annotix, CVAT, and Label Studio, in which three evaluators annotated 60 synthetic images across bounding box and mask tasks, analyzed with Kruskal–Wallis and Dunn–Bonferroni post hoc tests. Third, a heuristic usability evaluation assessed standardized tasks on real medical images (retinal fundus and otoscopic images). Results indicate that Annotix achieves annotation efficiency competitive with established tools while offering broader integrated coverage, including end-to-end model training and serverless peer-to-peer collaboration. Annotix is freely available under the MIT license and targets privacy-sensitive domains such as medical imaging and ecological monitoring.

Computers doi: 10.3390/computers15080495

Authors: Evgeniya Ishchukova Faezeh Sadat Sajadi Sergei Petrenko Alexey Petrenko Alexey Nekrasov

The harvest now, decrypt later (HNDL) attack is an attack that collects encrypted information now and decrypts it later after the arrival of a quantum computer that can perform some cryptographic operations. Time of exposure is data retention and not Q-Day, so the threat is imminent, but it is not reflected in the Common Vulnerability Scoring System (CVSS). We present the Quantum Risk Index (QRI), which is based on CVSS base severity, the Quantum Factor (QF, vulnerability to Shor’s or Grover’s algorithm), and the HNDL Score (HS, susceptibility to a harvest-and-store adversary). We computed the QRI for 78,587 CVEs that were cataloged in the NIST National Vulnerability Database between January 2016 and the first quarter of 2026 and cross-checked it with the CISA Known Exploited Vulnerabilities (KEV) list. The number of CVEs related to cryptography increased by a CAGR of 11.0%, while the number of Shor-vulnerable CVEs increased at a CAGR of 8.8%. The 437 KEV-matched CVEs carry a mean QRI of 13.05, against 10.31 for the non-KEV remainder—a 26.6% separation (p = 2.17 × 10−63)—and Shor-vulnerable CVEs appear in the KEV list at 1.52 times the baseline rate (p = 0.002). This separation is valid for each of the tested settings of QF and HS. This is the first study to apply a quantum-adjusted vulnerability score to a government’s in-the-wild vulnerability reporting database at the CVE scale, providing a repeatable foundation for quantum-aware vulnerability triage.

Computers doi: 10.3390/computers15080494

Authors: Asaeyl Alahmadi Salma Elhag Maram Meccawy

The transportation sector is a primary focus for Artificial Intelligence (AI) innovation because of ongoing technological progress. The sector faces growing challenges due to urbanization and congestion. Cities are considering electric and Autonomous Vehicles (AVs) to be promising future solutions because of their potential to enhance efficiency and reduce emissions, thereby helping us to achieve sustainability goals and drive the digital transformation. However, adopting these technologies requires evaluating their real-world deployment readiness. This research is motivated by the need to address this gap. The research develops and validates a smart framework through a technology readiness assessment methodology to evaluate AV readiness, combining rule-based reasoning with a weighted scoring and threshold-based classification model, along with Large Language Model (LLM) analysis. PESTEL and SWOT analyses were conducted to understand the current state of AV readiness in Saudi Arabia. The system was validated against a benchmark dataset derived from published sources of expert assessments across 20 countries. The system achieved a 97.1% agreement rate and a 0.44 Mean Absolute Error (MAE). The robustness of the weighted aggregation model was demonstrated through a Monte Carlo simulation, achieving a 92.1% stability rate when dealing with uncertain conditions. The framework proved its real-world value through a Saudi Arabian case study with an overall predicted TRL at 7–8, identifying social acceptance and regulation as a priority for development. These findings will advance AI adoption in the transport sector by providing a multidimensional framework to evaluate the readiness for deploying AVs.

Computers doi: 10.3390/computers15080493

Authors: Meng Yang Zhonghao Zhao Ning Chen Pengpeng Zhang

In the field of coal mine safety, traditional large language models (LLMs) face issues such as poor knowledge timeliness, lack of traceable evidence, and susceptibility to hallucinations when handling complex knowledge-intensive tasks. To align with the human-centric principles of Industry 5.0 and meet the strict compliance requirements of Industrial Regulatory Technologies (RegTechs), this study aims to enhance the performance of a Retrieval-Augmented Generation (RAG)-based coal mine safety compliance system. A comprehensive coal mine knowledge dataset (CMK) is constructed to serve as a dynamic RegTech knowledge base. Based on the RAG framework, a two-stage retrieval structure is designed and optimized, transforming the generative AI into a safe, constrained reasoning core. Also, a hard-negative ranking dataset (CMK-R) is developed to support the evaluation of the reranking model and mitigate the risk of semantic hallucinations. The proposed approach adopts Dmeta-embedding-base and BGE-rerank-base as backbone models and applies fine-tuning, teacher–student structured distillation, and Matryoshka Representation Learning (MRL) to improve efficiency and reduce energy consumption. Experimental results demonstrate that the proposed CMS-base model achieves FAHR, MRR@10, and mAP scores of 93.48, 95.59, and 95.63, respectively, indicating that the system can retrieve the correct safety regulation or operational clause at the top ranks with high reliability. Furthermore, the lightweight variant reduces inference time by over 60%, and accuracy degradation remains below 1% even when the vector dimension is compressed to 64, significantly facilitating Edge Intelligence deployments. This robust, human-centric framework provides a scalable industrial AI solution for real-time safety compliance and decision support.

Computers doi: 10.3390/computers15080492

Authors: Zhuoran Li Yang Wang Mengqing Yan Fan Zhou Yongxin Feng

Automatic Modulation Classification (AMC) aims to automatically identify modulation types based on the features of received signals, and acts as a vital technique for spectrum sensing, cognitive radio and electronic countermeasures. However, existing methods generally overlook the issue that modulation signals with similar characteristics are prone to confusion. In particular, under strong interference conditions, feature distributions become blurred and class boundaries tend to overlap, further exacerbating misclassification among modulation types with similar characteristics, thereby limiting improvements in classification accuracy and model robustness. To address this challenge, a modulation classification method based on fuzzy sample feature enhancement is proposed. Specifically, a modulation class entropy constraint and a fuzzy sample feature enhancement constraint are introduced to establish a Multi-scale Fuzzy Sample Feature Enhancement Framework (MTFSFEF). Through fuzzy sample selection and feature representation refinement, the proposed framework effectively mitigates feature overlap among fuzzy samples and enhances inter-class separability and discriminability. For SNR≥0dB, MTFSFEF delivers superior average classification accuracies across all three benchmark datasets: 92.82% on RML2016.10a, over 93.43% on RML2016.10b, and exceeding 92.78% on RML2018.01a, outperforming existing methods by up to 2.68%, 2.89%, and 2.73%, respectively.

Computers doi: 10.3390/computers15080491

Authors: Anna Kandylidou Georgios Spanos Charalampos Karagiannidis Filippos Vlachos Alexandros Nizamis Konstantinos Votis

The integration of virtual reality (VR) and music is gaining attention in therapeutic and rehabilitative fields for its potential role in health-related interventions. This scoping review maps how music and VR have been combined across health contexts. Using databases such as PubMed and Scopus, peer-reviewed studies published between January 2009 and 15 February 2024 were analyzed following established review reporting procedures. Forty-five studies were reviewed, covering diverse patient populations, including those with neurological impairments and emotional disorders. The mapped literature reports outcomes related to patient engagement, motor function, stress reduction, emotional satisfaction, feasibility, and usability. However, because intervention methods and patient needs vary, and because the included studies are heterogeneous in design, population, and outcome measures, the findings should be interpreted as evidence mapping rather than definitive evidence of clinical effectiveness. By mapping the current evidence, this review may support future research and clinical practice by helping to refine VR and music interventions for specific health contexts, patient populations, and therapeutic goals.

Computers doi: 10.3390/computers15080490

Authors: Samuel Simbarashe Furusa Mampilo Phahlane

The adoption of e-Government projects by African governments is increasing, but despite their implementation, high rates of failure and abandonment persist, resulting in significant financial losses and few improvements in the provision of public services. One significant challenge is that such failures are mostly associated with technical/infrastructural factors, which means that other underlying factors in the process of ICT projects’ failure and success remain underexplored, including the interaction between ICT governance and environmental factors. This research focuses on the interaction between ICT governance and the environment and its effects on ICT projects. Using the PRISMA model, the literature review identified 42 of 90 articles and employed thematic analysis for theme categorisation. The findings identify four interdependent dimensions of failure: contextual pressures, structural governance weaknesses, rigid process mechanisms, and relational misalignment. These factors interact in a sequential cascade of Context > Structure > Process > Relational, whereby institutional conditions reconfigure governance into a compliance-based model, resulting in reduced flexibility and coordination. This research builds an ICT governance failure model with multiple dimensions and defines the concept of a Governance–Reality Gap. The results show that achieving better ICT project performance requires the concurrent consideration of ICT governance structures and institutional environments.

Computers doi: 10.3390/computers15080489

Authors: Tetiana Babenko Kateryna Kolesnikova Askar Syssoyev Yelizaveta Vitulyova

Public-source evidence is widely used to assess exposed IoT and IIoT systems, yet its survival across successive stages of non-intrusive inference is rarely measured. This study examines how far an observer confined to public data can follow an evidence chain from a reachable host toward vulnerability and adversary-technique interpretation without privileged access, exploitation, or direct target interaction. Using a fixed, non-representative population of 227 public hosts constructed from IoT-23 network traffic, we measured evidence survival from surface observability through service and product evidence, CPE acceptance, and CVE resolution. The reported figures describe this constructed population rather than internet-facing IoT systems at large. The largest loss came first: only 58 hosts were externally observable, making observability the dominant constraint within this setting. A second source mainly deepened evidence for already visible hosts rather than recovering the blind region. Severity and exploitation-likelihood evidence survived at the CVE level, but technique-level interpretation was not pursued beyond the predefined external-observer model. We conclude that, within this measurement setting, external evidence-to-risk inference is bounded chiefly by observability and that distinct evidence layers should be measured separately rather than merged into a single risk signal.

Computers doi: 10.3390/computers15080488

Authors: Tao Li Ruihang Li Huangnan Zheng Heng Chen Kehan Wang Wangliang Guo Hong Li Shijian Li Zhijie Pan

Automated 3D city generation, focusing on road networks and building layouts, is in high demand for applications in urban planning, analysis, and simulations. The surge in deep generative models has facilitated automated design in recent years. However, the lack of high-quality datasets and benchmarks hinders the progress of these data-driven methods in generating city configurations. To fill this gap, this study introduces a multimodal dataset designed for the controllable generation of road networks and building layouts (named RoBus), whose public project repository provides release materials, and constitutes a large-scale resource in the field of generative city design. The RoBus dataset comprises aligned images, graphics, labels, and texts, with 72,400 paired samples that cover around 80,000 km2 globally. Besides utilizing prevalent generative models, we also introduce baseline models that leverage the multimodal features of RoBus. The experiments establish the dataset’s usability while revealing complementary trade-offs rather than uniform superiority. ControlNet obtains the lowest road network FID (20.78), whereas our topology-aware road baseline obtains the highest traffic-convenience score (0.83) at the cost of lower fidelity and diversity. For building layouts, our multimodal baseline reduces FID to 17.42 and building-density Wasserstein distance from 6.12 to 3.37 but produces lower diversity and a higher invalid-sample rate than the strongest comparison methods.

Computers doi: 10.3390/computers15080487

Authors: Zhiwei Yu Chuang Wang Min Xu Yuxiang Huang Kaiping Tian Yunchuan Qin

As Large Language Models integrate into Cyber–physical system like smart grids, their semantic vulnerabilities pose severe threats to physical infrastructure. Traditional digital domain alignment evaluations fail to capture these concrete physical risks. To address this, this paper proposes a transdomain dynamic quantitative evaluation framework driven by statistical mechanics and Monte Carlo simulations. Based on national grid standards, we construct a high fidelity adversarial dataset to test advanced structural transformation and logic driven coercion attacks. Furthermore, we introduce a Hazard Index (H) mapped to physical topologies and an entropy driven dynamic defense threshold (Ω). Through 10,000 concurrent simulations, our results reveal that traditional attack success metrics suffer from severe spurious positives. We find that high structural entropy injections drastically squeeze the attentional computing power of models, triggering a precipitous collapse of defense thresholds, whereas massive parameter scales provide critical attentional redundancy capacity to suppress physical penetration probabilities (P%). Consequently, static semantic filtering is fundamentally inadequate for industrial security. This study establishes a rigorous baseline for quantifying cascading physical risks, providing a solid theoretical foundation for the secure deployment of Artificial General Intelligence in future industrial internets.

Computers doi: 10.3390/computers15080486

Authors: Kanwal Naz Imran Shafi M. Zakria Mehmood Abdul Saboor Khan Jamil Ahmad

Maintaining code quality during software maintenance is a persistent challenge because bug-fix activities frequently introduce new code smells, accelerating long-term technical debt. Although bug-report classification has been applied to severity prediction and developer triaging, its use for proactively distinguishing Code-Quality-Impacting Bugs (CQIBs) from Non-Code-Quality-Impacting Bugs (NQIBs) remains underexplored. This paper proposes a transformer-based classification framework that fine-tunes CodeBERT on a balanced dataset of 25,000+ bug-report segments drawn from four Apache projects (Camel, CloudStack, Geode, and HBase). We evaluate two training strategies sequential transfer learning and individual fine-tuning and compare both against the published CNN-based baselines and against BERT, RoBERTa, and DeBERTa. Under 5-fold cross-validation, individual CodeBERT fine-tuning achieves weighted-F1 scores of 0.924, 0.866, 0.858, and 0.845 on Camel, CloudStack, Geode, and HBase, respectively, for an average of 0.873. Sequential transfer learning reaches a mean accuracy of 89.8% and a peak accuracy of 92.4% on Camel. CodeBERT is the best-performing transformer in this comparison, exceeding the average weighted-F1 of BERT (0.790), RoBERTa (0.753), and DeBERTa (0.742). Rigorous preprocessing, class balancing within the training folds, and stratified evaluation contribute substantially to these results.

Computers doi: 10.3390/computers15080485

Authors: Wei Liu Xuguang Chen Rafiza Abdul Razak

Background: Preschool educational apps are child-facing systems, yet the label ‘educational’ provides little assurance of interactional or pedagogical quality. Objective: This scoping review mapped preschool app-quality criteria and integrated them in a child–computer interaction (CCI) synthesis. Methods: Five databases were searched for English-language reports published from 2014 to 26 June 2026. From 785 raw records, 229 were screened and 178 reports were sought. Full texts were obtained for 152 reports (85.4%); 26 were not retrieved and 17 were excluded. The final corpus comprised 103 Central A reports used for nine-dimension coding and 32 Contextual B reports used for interpretation. Initial domains, informed by the review questions and prior frameworks, were refined through pilot coding and applied using stricter app-specific rules. The first author coded the corpus; the second checked flagged, borderline, revised, and sampled cases, with disagreements resolved by consensus. Results: Explicit reporting was most frequent for engagement, interaction and feedback (88/103), pedagogical design (84/103), developmental appropriateness (74/103), evidence and evaluation transparency (69/103), and content and curriculum relevance (66/103). Conclusions: The nine dimensions integrate concerns addressed separately in existing approaches. They constitute a conceptual synthesis rather than a validated rubric or certification tool.

Computers doi: 10.3390/computers15080484

Authors: Pablo Vicente-Martínez Adrián Chust-Ros Emilio Soria-Olivas María Ángeles García-Escrivà Edu William-Secin Manuel Sánchez-Montañés

Operational planning for large-scale events involves high uncertainty, heterogeneous data sources, and complex resource allocation decisions. Although digital twins and simulation models can support this process, their adoption is often limited by the technical expertise required to configure scenarios and interpret outputs. This paper introduces a Conversational Digital Twin Framework (CDTF) that integrates generative AI with simulation-based decision support, enabling users to configure operational scenarios through natural language. The framework comprises four layers: data integration, a simulation engine combining machine learning and discrete-event simulation components, a conversational AI interface, and a visualization layer for reports and dashboards. It is validated through a TRL4 prototype for football stadium operational planning at the Gran Canaria Stadium, using historical match records, external contextual sources, and synthetically generated operational variables where detailed stadium data were unavailable. The validation focuses on architectural feasibility, workflow integration, conversational configuration, and operational viability. Results show that the agent recognized user intent in 93.3% of evaluated prompts, extracted all explicitly stated parameters in the tested cases, detected all evaluated invalid inputs, and completed the full simulation workflow in all tested scenarios. Average conversational responses took less than six seconds, while complete simulation workflows required approximately 1 min. These findings suggest that conversational interaction can facilitate the access to simulation-based planning and support interactive what-if analysis for event operations, while future work must validate predictive performance with real operational datasets.

Computers doi: 10.3390/computers15080483

Authors: Priska Steininger Manfred Pamsl Helmut Lindner Klaus Gebeshuber Patrick Deininger

Enterprises increasingly run GPU-bound workloads such as Large Language Model (LLM) inference inside virtualized infrastructure, yet practitioners have little systematic guidance on how the choice of virtualization platform affects GPU passthrough in configuration effort as much as performance. This study’s primary contribution is a structured, paired comparison of configuration complexity and performance across six platforms (Proxmox VM and LXC, native KVM, OpenStack VM and Zun, and Podman) on a single NVIDIA RTX 4500 Ada GPU, emphasizing the under-documented container paths (OpenStack Zun and Podman with the Container Device Interface). The platforms differ most in configuration complexity, where OpenStack, especially Zun, proved the most demanding and Proxmox and LXC the most straightforward. As a confirmatory baseline, for a small-model, low-concurrency vLLM workload that does not saturate the device, exclusive passthrough yields statistically equivalent throughput, duration, and inter-token latency across all platforms, within 2% of bare metal; this equivalence is bounded to the non-saturating regime tested and is not a general claim. Running two containers concurrently (each capped at 45% of GPU memory, compute time-sliced) doubled the offered load and raised aggregate throughput by 12–14%, consistent with idle GPU capacity but not isolating a GPU-sharing benefit.

Computers doi: 10.3390/computers15080482

Authors: Yongjian Wang Aibo Song

Trusted Data Spaces (TDSs) have emerged as the core infrastructure for secure, privacy-preserving data circulation across industries and jurisdictions. However, state-of-the-art TDS implementations suffer from centralized platform monopoly, rigid cross-border governance failure, unfair value distribution, and poor scalability for global-scale collaboration. This paper proposes DAO-TDS, a novel decentralized autonomous trusted data space paradigm that enables centerless, cryptography-governed, and value-closed-loop data circulation. We make three core contributions: (1) We formalize the first anti-monopoly, incentive-compatible game-theoretic model for distributed TDS governance, with rigorous provable security guarantees; (2) we design an original Proof of Data Contribution (PoDC) consensus mechanism and a post-quantum secure Crypto-DAO governance protocol, with formal security proofs under the Universal Composability (UC) framework; (3) we implement a full prototype of DAO-TDS and conduct comprehensive, reproducible evaluations, showing that it supports 10,000+ distributed nodes with >12,000 TPS and <2 s 99th-percentile confirmation latency, while delivering >80% of generated value to data contributors (vs. <50% in centralized platforms). While the proposed paradigm demonstrates strong performance and security guarantees, it still faces challenges in adaptive cross-jurisdictional compliance and lightweight edge node deployment, which require further investigation.

Computers doi: 10.3390/computers15080481

Authors: Dipayan Roy Abhinav Shukla Ram Krishna Akuli Ayush Kumar Agrawal Pitshou N. Bokoro Parul Dubey

The prediction of election outcomes is a critical area in political analysis and decision-making. Accurate forecasting models can significantly influence electoral strategies and policy formulation. Existing models, however, face challenges in handling complex, dynamic, and high-dimensional election data. This paper addresses these issues by utilizing the Election Polls Dataset, which includes structured and unstructured data from multiple polling agencies such as YouGov (U.K. and NY, USA), Morning Consult (Washington, DC, USA), and Harris Insights (Chicago, IL, USA). We propose a novel hybrid approach combining Transformer models with ensemble learning techniques, including Random Forest, XGBoost, and Gradient Boosting, to enhance prediction accuracy. The novelty of this approach lies in the integration of Transformer’s attention mechanism with ensemble methods, improving both prediction accuracy and model stability. The performance of the model is evaluated using metrics such as accuracy, F1-Score, precision, recall, and AUC-ROC. Experimental results show that the proposed model outperforms existing methods, achieving a 93.4% accuracy, surpassing the baseline model by 1.4%.

Computers doi: 10.3390/computers15080480

Authors: Aibol Tileukhan Gulmira Bekmanova Valentina Franzoni Alibek Barlybayev Lena Zhetkenbay Altynbek Sharipbay Zhanar Lamasheva Assel Omarbekova Aizhan Nazyrova

Indoor navigation remains challenging for students with visual impairments because GPS is unavailable indoors and building layouts are often complex. This paper presents a wearable marker-assisted navigation system integrating QR code localization, SSD MobileNet V3 obstacle detection, TFmini-S LiDAR ranging, A*-based dynamic route planning, and audio feedback on a Raspberry Pi 5. The main contribution is an analytical framework relating marker spacing to predicted localization uncertainty and defining a latency budget for obstacle warnings. A confidence-weighted sensor-fusion method is developed analytically but was not implemented in the evaluated prototype, in which the QR code, camera, and LiDAR channels operated independently. The proposed fusion method and the simulated multi-floor planning extension require further experimental validation. Controlled tests produced a mean positioning error below 1.2 m, a LiDAR ranging MAE of 8.3 cm, and an object-detection throughput of 6–9 FPS. A pilot field evaluation covered nine routes totalling 901 m across two buildings and included one participant with self-reported vision loss of approximately 95%. All route trials were completed, although some required researcher assistance. The system remains a proof of concept and has not yet been evaluated against a baseline or with a sufficiently large target-user sample.

Computers doi: 10.3390/computers15080479

Authors: Jyotiprakash Mishra Sanjay K. Sahay Swati Mishra Aman Pathak

Microarchitectural side-channels leak data through caches, branch predictors, store buffers, and speculative execution. Evaluating defenses at cycle-model fidelity has forced a choice between functional tools (Spike, QEMU) and gem5’s hours-long Linux boots. ARCHER is a cycle-model-relative RV64IMAFDC RISC-V emulator with pluggable superscalar out-of-order execution, coherent caches, simultaneous multithreading, 1 to 16 harts, and thirteen speculation policies: an unrestricted baseline plus five classical defenses and seven new defenses spanning issue-gate, predictive-redirect, and selective-cleanup mechanism families. It boots Linux 6.6 in under three minutes, runs 3.7× faster (geometric mean) than gem5’s DerivO3CPU on head-to-head bare-metal microbench, matches gem5 within ±20% on 4 of 10 workloads, matches Chipyard’s fab-ready RTL to a 9.4% mean cycle-count deviation under two fitted match-configurations, and produces bit-identical architectural output to QEMU at a median 88× host-throughput advantage over Chipyard’s Verilator flow. Every published Spectre-family defense drives leakage to zero at sub-0.3% instructions-per-cycle (IPC) loss on a full Linux boot, and each new policy is validated on the attack surface its mechanism defends. The Adaptive Memory Encryption Scheme (AMES) adds a bus-layer engine with four authenticated ciphers (each validated bit-exact against its published specification) and per-leaf mask-XOR for Differential Power Analysis (DPA) hardening; a first-order Correlation Power Analysis consistency check confirms the direction of the theoretical bound under the shipped leakage model. Together, ARCHER and AMES provide a single INI-configurable environment for cycle-model side-channel evaluation in minutes rather than hours.

Computers doi: 10.3390/computers15080478

Authors: Mostafa Aboulnour Salem

Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG)-based hybrid AI systems are increasingly transforming higher education, yet it remains unclear whether they are adopted through similar or distinct technology adoption pathways. This study compares the adoption of these two AI paradigms among postgraduate students using the Unified Theory of Acceptance and Use of Technology (UTAUT). It proposes the Adaptive Dual-AI Model (ADAM), which extends UTAUT by examining architecture-sensitive AI adoption within a unified framework. The model incorporates Technology Readiness (TR) and AI Awareness (AIA) as mediating variables and generational differences (Gen Z and Gen Y) as moderating factors. Data were collected from 639 postgraduate students enrolled in Saudi universities, of which 619 valid responses were retained after data screening. Partial Least Squares Structural Equation Modelling (PLS-SEM) was employed to evaluate the proposed model. The results indicate that the two AI paradigms follow distinct technology adoption pathways. Behavioural intention toward standalone LLMs was primarily associated with Performance Expectancy (PE) and Effort Expectancy (EE), whereas the adoption of RAG-based hybrid AI systems was more strongly associated with indirect relationships involving Technology Readiness and AI Awareness. Multi-group analysis further revealed that Gen Z learners exhibited stronger associations with LLM adoption, whereas Gen Y learners demonstrated stronger relationships involving RAG-based hybrid AI systems. These findings support the proposed architecture-sensitive perspective of ADAM, suggesting that differences in AI architecture are associated with distinct technology adoption patterns. From a practical perspective, the findings provide guidance for higher education institutions in selecting complementary AI technologies according to learning objectives, learner characteristics, and evidence requirements. More broadly, the study contributes to technology adoption research by extending UTAUT to heterogeneous AI ecosystems and offers practical insights for designing adaptive, personalised, and evidence-aware learning environments.

Computers doi: 10.3390/computers15080477

Authors: André Costa João Miranda João Mirra Nuno Dinis Luís Romero Pedro Miguel Faria

The increasing digitalization of industrial environments has led to the emergence of new paradigms such as Digital Twins (DTs), which enable real-time connections between physical assets and their virtual counterparts. In this context, the Coffee Twin system was developed as a modular and controlled demonstrator to explore, present, and validate the integration of DT technology with Augmented Reality (AR), Virtual Reality (VR), and voice-based interaction. A coffee machine was selected as the case study due to its manageable complexity, low-risk operation, and potential to illustrate concepts that can later be extended to more complex industrial systems. The system architecture is centered on a Raspberry Pi, which supports local data acquisition from temperature and humidity sensors, mechanical actuation through a servo motor, and remote control via a smart switch. To interact with the system, three software applications were developed: a Web application, an AR application for contextual visualization and remote interaction, and a VR application that recreates the operation of the machine within an immersive 3D environment. Beyond the technical implementation, the prototype was evaluated in demonstration and user-testing sessions in which 80 participants, mostly secondary-school pupils and university students, completed a coffee-making task across the three applications and then answered a System Usability Scale (SUS) questionnaire. The analysis of the 80 valid responses yielded a mean SUS score of 69.2 (standard deviation 15.9; 95% confidence interval [65.7, 72.8]), which corresponds to acceptable, slightly above-average usability and indicates that combining real-time sensing, remote actuation, and multimodal interaction in a single low-cost platform is perceived as usable across a heterogeneous, predominantly non-expert audience. The main contribution of this work is therefore not a new Digital Twin algorithm, but an accessible, modular, and reproducible integration of Web, AR, VR, and voice interaction around a physical asset, supported by usability evidence. Overall, Coffee Twin shows how a low-cost and multimodal DT demonstrator can support the communication, experimentation, and early exploration of Industry 4.0 concepts in educational and pre-industrial settings.

Computers doi: 10.3390/computers15080476

Authors: Samuel Simbarashe Furusa Mampilo Phahlane

ICT governance is becoming indispensable for the planning and delivery of public sector services as more digital platforms are utilised in these processes. This research presents an overview and interpretation of the ICT governance adoption process in Africa’s public sector organisations using McClelland’s Need Theory. This systematic review used the SPAR-4-SLR methodology and the ADO framework. A collection of scholarly articles and grey literature on ICT governance between 2010 and 2025 was obtained through searches of the Scopus and Web of Science databases. Based on the results, it is apparent that ICT governance adoption in Africa is institutionally uneven and cannot be explained solely by technological models. While factors such as governance systems, regulations, and alignment contribute to institutional adoption of governance, relational factors such as stable leadership, cooperation, accountability, and trust seem to carry more weight. From the review, it becomes clear that, in some instances, governance structures are used symbolically to demonstrate adherence and legitimacy without operational integration within the institutions. This study concludes that ICT governance reforms in the African public sector must strike a balance among authority, cooperation, and governance performance to create value for society.

Computers doi: 10.3390/computers15080475

Authors: M. Riduan Abid

Artificial Intelligence (AI) and Edge Computing (EC) have emerged as two of the most transformative forces shaping contemporary computing systems [...]

Computers doi: 10.3390/computers15080474

Authors: Husein Mauladdawilah Mohammed Balfaqih Zain Balfagih Aimad El Habti María del Carmen Pegalajar Eulalia Jadraque Gago

Accurate 24 h photovoltaic (PV) power forecasting is essential for day-ahead scheduling, storage operation, reserve planning, and market participation. However, published deep learning comparisons are often difficult to reproduce and interpret because they use inconsistent datasets, forecasting horizons, baselines, evaluation metrics, and leakage-control procedures. From a software engineering perspective, this limits the trustworthiness, comparability, and practical adoption of AI-based forecasting systems. This paper presents a controlled and reproducible benchmarking framework for evaluating AI-driven forecasting software. The framework is applied to nine deep learning architectures, three non-deep learning reference models, and two persistence baselines for hourly PV-power forecasting at a 350 kWp rooftop installation near Edinburgh, Scotland. All models were evaluated under a consistent experimental protocol, including the same chronological train–validation–test split, a 32-feature meteorological and solar-geometry input set, a 24-step forecasting horizon, capacity-normalised mean absolute error (NMAE), and Bayesian hyperparameter optimisation. The results show that TCN-LSTM achieved the best aggregate H24 performance with 7.22% NMAE, narrowly outperforming CPWformer-DEC at 7.28% and CT-PatchTST at 7.31%. LightGBM ranked fourth at 7.35% with fixed hyperparameters, outperforming six of the nine deep learning models. The top three models differed by only 0.09 percentage points, indicating that architectural superiority cannot be established reliably without significance testing and operational diagnostics. Per-horizon analysis showed that CT-PatchTST and S-Mamba performed best at the nearest forecast steps, whereas TCN-LSTM provided the most stable far-horizon profile. Peak-power diagnostics further revealed that aggregate NMAE can mask operational shortcomings, as Naive Persistence outperformed all deep learning models in high-output peak detection. The findings highlight the importance of reproducible benchmarking, leakage safeguards, horizon-aware evaluation, and operationally meaningful diagnostics in trustworthy AI software evaluation. The novelty of this work lies not in proposing a new architecture but in a controlled, reproducible framework that benchmarks fourteen forecasters under identical conditions, with explicit leakage safeguards, per-horizon reporting, and operationally meaningful peak diagnostics, enabling claims of architectural superiority to be made trustworthy rather than merely favourable. Architecture selection for PV forecasting should therefore consider not only aggregate accuracy but also reliability, interpretability of evaluation outcomes, and deployment-relevant performance behaviour.

Computers doi: 10.3390/computers15080473

Authors: Noora Shihab Ahmed Farnaz Mahan Jaber Karimpour

Accurate classification and efficient retrieval of histopathological images are essential for the diagnosis of lung adenocarcinoma (LUAD). Existing deep learning approaches for Content-Based Histopathological Image Retrieval (CBHIR) typically generate single-scale embeddings, missing the richer spatial context from earlier network stages. We propose a unified framework composed of: (1) a ConvNeXt V2 backbone with an integrated Convolutional Block Attention Module (CBAM) for multi-scale feature extraction, (2) an Adaptive Weighted Fusion Neck with learnable softmax-normalized weights, and (3) a novel Hybrid Representation Head producing an 18,496-dimensional descriptor by concatenating global, spatial, and attention-weighted features. Evaluated on the WSSS4LUAD dataset (10,087 patches, four tissue classes), our model achieves 85.03% accuracy (5-fold CV: 83.35 ± 0.93%), F1-score of 0.8116, mean Average Precision (MAP) of 0.8323 for retrieval, and an Expected Calibration Error (ECE) of 0.0378. Ablation experiments confirm that all proposed modules contribute positively, with the Attention Branch being the most impactful (Δ = −2.13%). The framework further provides Gradient-weighted Class Activation Mapping (Grad-CAM) explainability for clinical interpretability.

Computers doi: 10.3390/computers15080472

Authors: Jianbo Zhao Kainan Zhang Zilong Yuan Weimin Wang Fei He

The efficiency of ammunition support operations in the aircraft carrier intermediate ordnance staging deck is critical to sortie generation rates in naval aviation, yet joint scheduling and personnel allocation in this multistage, resource-constrained environment remains a challenging bi-objective optimization problem. This study develops a framework integrating an improved Nondominated Sorting Genetic Algorithm II (NSGA-II) with a marginal-benefit-based iterative feedback mechanism. The intermediate ordnance staging deck support process is decomposed into individual ammunition processing stations and formulated as a processflow model incorporating operation sequencing and personnel specialization constraints. A constraint decision model then dynamically reconciles the minimization of total makespan and personnel workload equilibrium through iterative marginal-benefit comparison across support teams. The NSGA-II is enhanced with an adaptive crossover-mutation mechanism and an improved elitism preservation strategy to strengthen global search capability. Validation on a typical carrier intermediate ordnance staging deck scenario demonstrates that the improved NSGA-II outperforms the conventional NSGA-II in convergence speed and Pareto front quality. Under the optimized configuration, the total makespan remains 3600 s with a workload balance metric of 1075 even as ammunition quantity doubles from two to four units. The proposed framework offers practical decision support for carrier ammunition operations and extends to other resource-constrained multi-objective scheduling domains.

Computers doi: 10.3390/computers15080471

Authors: Aymen Akremi

Large Language Models (LLMs) have significant potential in regulated domains such as law, healthcare, and compliance, where users need help interpreting complex rules and documents. However, these domains also make the risks of Large Language Models especially serious: a system may hallucinate legal authority, rely on outdated rules, mix jurisdictions, or expose sensitive information. Retrieval-Augmented Generation (RAG) reduces these risks by grounding the model’s answer in a curated document corpus, but standard RAG still does not guarantee that the retrieved sources are legally valid, up to date, applicable to the correct jurisdiction, or safe to use. In this paper, we present an ontology-governed approach to legal RAG. The central idea is to use a legal ontology not merely as background knowledge, but as an active control layer. Before retrieval, the ontology filters legal sources by jurisdiction, topic, lifecycle status, and temporal validity. After generation, validation rules check whether the answer is supported by approved evidence, cites valid legal sources, respects jurisdictional boundaries, and avoids unsafe or privacy-violating content. The system also records retrieval, validation, and response-generation steps in an audit trail to support later review. In this way, the proposed Legal Rule Auditor extends Graph RAG from a retrieval-enhancement technique into a governance architecture for legal question answering. Its goal is not simply to improve answer relevance, but to ensure that answers are legally grounded, trusted, current, jurisdictionally appropriate, and traceable.

Computers doi: 10.3390/computers15080470

Authors: Tehmina Amjad

This special issue presents a notably broad view of contemporary data mining and its applications [...]