Computers

37 pages, 3573 KB

Open AccessArticle

A Multi-Stage Deep Learning Framework for Automated Brain Tumor Diagnosis and Clinical Report Generation

by Mohamed Eassa, Nagwa Yaseen Hegazy and Hussam Elbehiery

Computers 2026, 15(8), 519; https://doi.org/10.3390/computers15080519 - 11 Aug 2026

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 [...] Read more.

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. Full article

(This article belongs to the Section AI-Driven Innovations)

► Show Figures

Figure 1

44 pages, 508 KB

Open AccessSystematic Review

Bridging Two Worlds: Sensor Technologies and AI for Fruit Detection in Latin America and Beyond—A Scoping Review

by Franklin Parrales-Bravo, Joan Gracia-Chinga, Janio Jadán-Guerrero, Leonel Vasquez-Cevallos, Lorenzo Cevallos-Torres and Leili Lopezdominguez-Rivas

Computers 2026, 15(8), 518; https://doi.org/10.3390/computers15080518 - 10 Aug 2026

Abstract

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 [...] Read more.

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. Full article

(This article belongs to the Special Issue Intelligent Computing and Sensing Systems for Sustainable Precision Agriculture)

► Show Figures

Figure 1

37 pages, 2443 KB

Open AccessArticle

A Collaborative Decision-Making Model Based on Blockchain-Driven Adaptive Consensus for Public Opinion Event Response

by Yuetong Chen, Yumei Wang, Yufu Ning, Fengming Liu and Mingrui Zhou

Computers 2026, 15(8), 517; https://doi.org/10.3390/computers15080517 - 10 Aug 2026

Abstract

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 [...] Read more.

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. Full article

(This article belongs to the Topic Decision Science Applications and Models (DSAM))

27 pages, 26623 KB

Open AccessArticle

Beep and Show, Don’t Tell: Multimodal Feedback Strategies for Driver-Automation Coordination in Critical Driving Situations

by Stefan Reitmann, Tsvetomila Mihaylova, Dionysios Kritharoulas, Eelis Peltola, Elin A. Topp and Ville Kyrki

Computers 2026, 15(8), 516; https://doi.org/10.3390/computers15080516 - 8 Aug 2026

Abstract

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: [...] Read more.

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. Full article

(This article belongs to the Special Issue Advanced Human–Robot Interaction 2026)

► Show Figures

Figure 1

21 pages, 1055 KB

Open AccessArticle

Beyond Traditional Metrics: Toward a Multifactorial Model for Measuring Productivity in Agile Software Teams

by Marcela Guerrero-Calvache, Giovanni Hernández and María Clara Gómez-Álvarez

Computers 2026, 15(8), 515; https://doi.org/10.3390/computers15080515 - 8 Aug 2026

Abstract

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, [...] Read more.

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. Full article

(This article belongs to the Special Issue Advanced Software Engineering in the Age of Artificial Intelligence: Best Practices, Challenges, and Opportunities)

► Show Figures

Figure 1

21 pages, 2796 KB

Open AccessArticle

FD-GhostFaceNet: Frequency-Decoupled Ghost Modules for Lightweight Face Recognition

by Abdalbasit Qadir, Bryar A. Hassan and Hozan Khalid

Computers 2026, 15(8), 514; https://doi.org/10.3390/computers15080514 - 7 Aug 2026

Abstract

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 [...] Read more.

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. Full article

(This article belongs to the Section AI-Driven Innovations)

► Show Figures

Figure 1

24 pages, 27960 KB

Open AccessArticle

A Lightweight Shallow CNN-Based Approach for Ring-Neck Disease Detection in Avocado Fruits

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

Computers 2026, 15(8), 513; https://doi.org/10.3390/computers15080513 - 7 Aug 2026

Abstract

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 [...] Read more.

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. Full article

► Show Figures

Figure 1

25 pages, 1597 KB

Open AccessArticle

From Classical to Deep Learning: A Hybrid CNN–Ensemble Framework for Intrusion Detection in Internet of Medical Things

by Faris Kateb, Owais Khan and Fazal Qudus Khan

Computers 2026, 15(8), 512; https://doi.org/10.3390/computers15080512 - 7 Aug 2026

Abstract

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 [...] Read more.

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. Full article

(This article belongs to the Special Issue IoT: Security, Privacy and Best Practices (3rd Edition))

► Show Figures

Figure 1

13 pages, 8118 KB

Open AccessOpinion

Digital White Spaces: A Cyberpsychology-Informed Framework to Mobile Phone Addiction

by Leandros A. Maglaras, Christina Kyritsi, Helge Janicke and Konstantinos Karantzalos

Computers 2026, 15(8), 511; https://doi.org/10.3390/computers15080511 - 7 Aug 2026

Abstract

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), [...] Read more.

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. Full article

(This article belongs to the Section ICT Infrastructures for Cybersecurity)

► Show Figures

Figure 1

40 pages, 2833 KB

Open AccessArticle

Agentic Shadow Infrastructure: How AI Supply-Chain Drift Creates Unmanaged Enterprise Infrastructure

by Robert Campbell

Computers 2026, 15(8), 510; https://doi.org/10.3390/computers15080510 - 6 Aug 2026

Abstract

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 [...] Read more.

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. Full article

(This article belongs to the Section ICT Infrastructures for Cybersecurity)

► Show Figures

Figure 1

27 pages, 872 KB

Open AccessArticle

Compositional Formal Verification of Anti-Lock Braking System Neural Controllers

by Huixing Fang

Computers 2026, 15(8), 509; https://doi.org/10.3390/computers15080509 - 6 Aug 2026

Abstract

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 [...] Read more.

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 \geq 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. Full article

► Show Figures

Graphical abstract

29 pages, 1449 KB

Open AccessArticle

Solving Flow-Shop Scheduling Problems with Random Machine Breakdown and Limited Buffer Using a Pigeon-Inspired Hybrid Artificial Bee Colony Algorithm

by Mariappan Kadarkarainadar Marichelvam and Mariappan Geetha

Computers 2026, 15(8), 508; https://doi.org/10.3390/computers15080508 - 6 Aug 2026

Abstract

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 [...] Read more.

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. Full article

(This article belongs to the Special Issue Operations Research: Trends and Applications)

► Show Figures

Graphical abstract

19 pages, 460 KB

Open AccessArticle

AI as Cognitive Complement or Replacement? Perceived AI Role and Cognitive Independence Among University Students

by Dalma Lilla Dominek, Vanessza Kapusi, Szabolcs Ceglédi and Zoltán Szűts

Computers 2026, 15(8), 507; https://doi.org/10.3390/computers15080507 - 6 Aug 2026

Abstract

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 [...] Read more.

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. Full article

(This article belongs to the Special Issue The Digital Transformation of Education: Trends, Technologies, and Responsible Innovation)

► Show Figures

Graphical abstract

27 pages, 1346 KB

Open AccessSystematic Review

Explainable Artificial Intelligence for Tabular Data in Healthcare: A Systematic Review of Methods, Evaluation, and Applications

by Angelower Santana-Velásquez and Maria Bernarda Salazar-Sánchez

Computers 2026, 15(8), 506; https://doi.org/10.3390/computers15080506 - 6 Aug 2026

Abstract

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 [...] Read more.

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. Full article

(This article belongs to the Special Issue Explainable Artificial Intelligence for Signal Processing and Recognition)

► Show Figures

Figure 1

28 pages, 3358 KB

Open AccessArticle

A Trust-Aware Extension to a Reinforcement Learning Hyper-Heuristic Framework for Multi-Objective Scientific Workflow Scheduling

by Hadeel Amjed Saeed, Sufyan T. Faraj Al-Janabi, Esam Taha Yassen and Omar A. Aldhaibani

Computers 2026, 15(8), 505; https://doi.org/10.3390/computers15080505 - 5 Aug 2026

Abstract

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 [...] Read more.

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. Full article

► Show Figures

Figure 1

26 pages, 3518 KB

Open AccessArticle

Cost-Aware Android Malware Detection Using an Early-Warning Behaviour Score

by Ali Fenjan, Mohammed Almulla and Jalil Md. Desa

Computers 2026, 15(8), 504; https://doi.org/10.3390/computers15080504 - 5 Aug 2026

Abstract

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 [...] Read more.

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. Full article

(This article belongs to the Topic Addressing Security Issues Related to Modern Software)

► Show Figures

Graphical abstract

24 pages, 28547 KB

Open AccessArticle

Project-Based Learning in Computer Engineering: Design and Implementation of a Smart Fisio System for Rehabilitation Training

by 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 and Brian Roberto Gómez-Martínez

Computers 2026, 15(8), 503; https://doi.org/10.3390/computers15080503 - 5 Aug 2026

Abstract

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 [...] Read more.

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). Full article

► Show Figures

Figure 1

22 pages, 7546 KB

Open AccessArticle

Attn-ChurnNet: A Transformer-Based Sequential Framework for Customer Churn Prediction in Subscription Platforms with Focal Loss Training and Conformal Uncertainty Quantification

by Didar Hossain, Mohiuddin Mehedi, Khandakar Rabbi Ahmed, Md Rafiul Mahmud, Mainul Islam Khan and Sakib Salam Jamee

Computers 2026, 15(8), 502; https://doi.org/10.3390/computers15080502 - 4 Aug 2026

Abstract

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 [...] Read more.

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;

T_{s}^{*}

= 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. Full article

(This article belongs to the Special Issue Advances in Explainable and Multimodal AI for Intelligent Systems and Medical Applications)

► Show Figures

Graphical abstract

28 pages, 784 KB

Open AccessArticle

Predictive Analytics in Cloud-Native Privilege-Escalation Detection: Enhancing Accuracy Through Temporal Graph Attention and Reinforcement Learning

by Md Nuruzzaman Pranto, Md Deluar Hossen, Mamunur R. Raja, Md Sharfuddin, Balayet Hossain and Khandakar Rabbi Ahmed

Computers 2026, 15(8), 501; https://doi.org/10.3390/computers15080501 - 3 Aug 2026

Abstract

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 [...] Read more.

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. Full article

(This article belongs to the Special Issue From 5G to 6G: Emerging Technologies in Wireless Networks)

► Show Figures

Graphical abstract

Journal Description

Computers

Latest Articles

Journal Menu

Journal Browser

Highly Accessed Articles

Latest Books

E-Mail Alert

News

Topics

Conferences

Special Issues

Further Information

Guidelines

MDPI Initiatives

Follow MDPI