Journal of Computing & Biomedical Informatics

Health-Aware Travel Recommendation Through Environmental Risk Profiling and Carbon-Aware Model Selection

Muhammad Ahmad Zia — 2026-08-14

The carbon footprint of frequent model retraining is a growing component of recommendation system operating cost, and travel recommendation systems rarely incorporate destination-level health information for users with chronic conditions. This paper presents a methodological framework that integrates both concerns at the feature engineering and model-selection stages of a tabular recommender pipeline. The construction draws on 980 real user preference profiles from the UCI Travel Review Ratings dataset, 81 globally distributed destinations enriched with elevation, climate, ultraviolet, and air-quality measurements retrieved at runtime from Open-Meteo, and condition-specific environmental thresholds grounded in clinical guidance for asthma and chronic obstructive pulmonary disease, cardiovascular disease, pregnancy, photosensitivity, and diabetes. Health conditions are sampled at WHO-aligned prevalence rates, and the recommendation label is constructed as a transparent weighted combination of preference match and health safety with additive Gaussian noise; the design discussion in Section 3.4 motivates this choice in the absence of a public dataset linking real user travel preferences with real health-condition records. Eleven machine learning models are evaluated across predictive accuracy, ranking quality, training time, energy consumption, and CO₂ emissions over three random seeds. XGBoost reaches the lowest RMSE of 0.266 and the highest NDCG@10 of 0.983 while emitting 0.007 g CO₂ per training run. The 200-tree Random Forest reaches a slightly worse RMSE of 0.287 while emitting 0.152 g, an emission ratio of approximately 22 with no measurable accuracy improvement. On the present benchmark, gradient boosting therefore Pareto-dominates Random Forest at every complexity level evaluated, a pattern consistent with the wider tabular-data literature. The health-aware feature pipeline shifts top-ten recommendations for non-healthy subpopulations away from destinations whose environmental profile exceeds condition-specific clinical thresholds, providing an implicit ablation against a preference-only baseline. The framework is released as an open-source replication package, and the present results should be read as a structured engineering proof of concept rather than a clinical validation.

A Deep Learning Approach for SMS Spam Detection Using Bert and Masked Language Models: A Case Study on Arabic and English

Musab Al-Ghadi — 2026-09-01

SMS spam poses a significant threat to the security and privacy of mobile users, causing financial fraud, phishing attacks, and diminishing trust in mobile communication systems. With multilingual and low-resource languages (like Arabic) with their linguistic complexity and dialectal variation, as well as the lack of labeled datasets for these languages, traditional spam detection methods are ineffective. This paper presents a novel multilingual SMS spam detection framework that leverages Bidirectional Encoder Representations from Transformers (BERT) with domain-adaptive Masked Language Modeling (MLM). Two models are created and tested: fine-tuned (FT) model on SMS corpus with labels and pre-trained (MLM-SMS) model on SMS corpus using MLM pretraining first followed by supervised fine-tuning. To evaluate the models, they are applied to the UCI English SMS Spam Collection dataset (5,574 messages) and a newly created Arabic dataset of 4,623 real and AI-generated SMS messages. Both models are trained using the multilingual variant of BERT, and share the same preprocessing pipeline and fixed sequence length of 128 tokens. The results demonstrate that the MLM enhanced model always works better than the baseline. It obtains 99.2% accuracy and 98.6% F1-scores on English data, whereas BERT-SMS gets 97.3% accuracy and 97.4% F1-scores. It achieves a higher accuracy of 98.8% and F1-score of 98.3% on the Arabic data compared with a baseline accuracy of 97.5% and F1-score of 96.6%. Another significant improvement in robustness is shown by the reduced number of false positives and false negatives, confirmed by the analysis of the confusion matrix. The proposed framework is scalable, language independent, and extremely effective for spam detection in real-time SMS in multilingual low resource and high resource settings.

Probabilistic and Information Theoretic Enabled Secret Key Generation in Wireless Ad Hoc Networks

Muhammad Sajid — 2026-08-14

This paper targets wireless ad hoc networks (WANETs) as the main domain and focuses on the sub-domain of secret key generation (SKG) from wireless channels. We consolidate probabilistic and information-theoretic foundations with deployable physical-layer techniques that exploit channel reciprocity, including RSS/CSI/phase features, UWB links, and MIMO-OFDM (plus long-range LoRa), and evaluate them using key rate, key-error-rate, and entropy. We then cover special methods that strengthen SKG pipelines: reciprocity calibration for TDD/FDD, adaptive quantization, code-based reconciliation (polar/lattice), privacy amplification via universal hashing, deep-learning aids for feature extraction and reconciliation, and IRS-assisted links to boost randomness under mobility. Building on 120 papers, we structure the literature into four 30-paper chunks with summaries (techniques, pros/cons, limitations, applications) and per-chunk best-use-case guidance, supported by explicit inclusion/exclusion criteria. Finally, we propose a practical SKG workflow for WANETs: (i) channel measurement & reciprocity calibration, (ii) adaptive quantization, (iii) information reconciliation with capacity-approaching codes, (iv) privacy amplification to bound leakage, and (v) scenario-specific deployment for IoT/WBAN/VANET. This yields a concise playbook that links rigorous secrecy guarantees to fieldable designs.

A Hybrid Ensemble Learning Framework for Educational data mining and Learning Analytics Research from an Interdisciplinary Perspective

Lu Li — 2026-09-01

The growing volume of educational data generated by Virtual Learning Environments (VLEs) presents unprecedented opportunities for data-driven student support. This paper proposes HELM a Hybrid Ensemble Learning framework with Multi-modal behavioral analytics designed to predict student academic performance and provide early at-risk identification using the Open University Learning Analytics Dataset (OULAD). HELM integrates five interdisciplinary components: (1) multi-source temporal feature engineering from VLE interaction logs, assessment records, and registration data; (2) unsupervised behavioral clustering to identify four distinct student archetypes; (3) a SMOTE-balanced stacking ensemble combining Random Forest, XGBoost, LightGBM, and Gradient Boosting with a Logistic Regression meta-learner; (4) an Early Warning System (EWS) evaluated across nine temporal prediction windows; and (5) SHAP-based explainability analysis. Experimental results on 32,593 student records demonstrate that HELM achieves an AUC-ROC of 0.9412, accuracy of 0.8876, and F1-score of 0.8901, outperforming all individual baseline models. The EWS achieves AUC > 0.85 as early as Day 30 of the module, enabling timely institutional intervention. Interdisciplinary analysis further reveals that socioeconomic deprivation (IMD band), prior education level, and VLE engagement regularity are the strongest predictors of student success. These findings offer actionable insights for educators, policymakers, and learning system designers.

Mediating Role of Quality Assurance on The Relationship Between TQM Practices and Education Performance in Higher Education Institutions of Pakistan

Baqir Nadeem Hashmi — 2026-08-05

Educational performance is a critical element that helps to assess how well a university is performing in terms of education, developing skills in her students, and bringing about significant societal changes. Prior literature reveals that quality assurance despite considered an integral component, has been neglected to examine between Total Quality Management (TQM) and education performance. The main objective of this research was to analyze the impact of TQM practices on educational performance, and to understand the role of quality assurance as a mediating factor between TQM practices and educational performance. To achieve this, a structured questionnaire was designed and distributed among the top management of universities of Pakistan to gather relevant information using convenience sampling technique. A total of 500 questionnaires were distributed, among them 264 were fully completed and returned, resulting in a response rate of 53%. The data was evaluated using the Partial Least Squares Structural Equation Modelling (PLS-SEM) approach. The results of the data analysis reveal that TQM practices are significantly related to educational performance, as well as to quality assurance. Furthermore, quality assurance was found to have a significant and positive impact on education performance. This research ultimately demonstrates that quality assurance serves as a significant mediator between TQM practices and education performance in the context of higher education institutions. The study implicates for the theory and practitioners that both TQM and quality assurance are pivotal antecedents to enhance education performance in Pakistani universities. Moreover, quality assurance plays a crucial role in developing an underlying mechanism with TQM practices to harness education performance in Pakistani universities.

Machine Learning Approaches for Chronic Pain Classification and Management: A Comparative Analysis Using the NHANES Dataset

Muhammad Umer Istiaq — 2026-09-01

Chronic pain affects over 30% of the global population, representing a major socio-economic and clinical burden. Current pain assessment is largely subjective, and machine learning (ML) offers a data-driven alternative for classifying chronic pain severity. This study evaluated five supervised ML algorithms (Logistic Regression, Random Forest, Support Vector Machine, XGBoost, and an Artificial Neural Network) for classifying chronic musculoskeletal pain severity (mild, moderate, severe) using pooled NHANES data (2015–2016 and 2017–2018), yielding an analytic sample of 9,971 adults, of whom 2,890 met a physician-diagnosed case definition based on arthritis and/or gout. Eight clinically relevant features spanning physiological, psychological, and behavioral domains were engineered, and models were evaluated using accuracy, AUC-ROC, macro F1-score, and Matthews correlation coefficient (MCC). XGBoost and Random Forest achieved the highest cross-validation accuracy (47.9% and 46.9%), significantly outperforming the ANN and simpler baselines, though the ANN achieved the highest macro F1-score and MCC on the held-out test set. Additional gradient-boosting variants, a deep neural network, and an explainability analysis were also evaluated as robustness checks, with LightGBM achieving the best overall held-out performance and the deep neural network underperforming the tree-based ensembles. Sleep disturbance, age, depression severity, and BMI emerged as the most consistently informative predictors, though all effect sizes were modest. These findings indicate that routinely collected demographic and clinical covariates provide limited discriminative power for fine-grained chronic pain severity classification, underscoring the need for richer feature sets in future ML-based pain assessment tools.

Gamification as a Means of Increasing Students' Motivation for Physical Activity

Yu Zhao — 2026-09-22

Sustaining students' motivation for physical activity is a central challenge in university physical education. Gamification addresses this challenge by structuring participation around clear goals, progress feedback, achievable challenges, rewards, and social interaction. Objective: This study examines the relationship between students’ exercise patterns and psychological well-being and uses this evidence to develop a mathematically specified gamification framework for strengthening motivation, engagement, and sustained participation in university physical activity. Methods: Data came from the 2021 Chinese General Social Survey. The analytic sample included 313 enrolled tertiary students aged 18-30 years. Exercise frequency ranged from never to daily, and elevated depressive mood was defined as feeling depressed or down at least sometimes during the previous four weeks. Binary and ordinal regression models adjusted for age, sex, household registration, socioeconomic position, and education type. In addition to the observational models, a mathematical framework is specified for evaluating gamified physical-education interventions through repeated measures of motivation and activity adherence. Results: Weekly-or-more exercise was reported by 65.8% of students, while 23.5% met the elevated depressive-mood threshold. Weekly-or-more exercise was not significantly associated with elevated depressive mood after adjustment (OR = 0.86, 95% CI 0.48-1.55, p = .615). The ordinal model likewise showed no dose-response association (OR = 0.95, 95% CI 0.77-1.17, p = .628). Conclusions: Exercise frequency alone did not distinguish students by depressive mood, indicating that frequency is an incomplete description of participation. For gamified physical education, motivation should be modeled directly through documented design features and repeated motivational outcomes rather than inferred from activity frequency. The CGSS does not measure gamification or exercise motivation, so the proposed framework requires controlled testing in university PE settings with longitudinal follow-up where possible.

Bin-Wise Quantized CatBoost for Energy-Efficient Drug-Review Rating Prediction in Biomedical E-Commerce

Muhammad Ahmad Zia — 2026-08-14

Artificial intelligence is increasingly applied on biomedical e-commerce platforms such as online drug-review aggregators, and machine learning provides most of its practical capabilities. Ma-chine learning models for predicting patient-generated ratings on such platforms are typically optimized for accuracy alone, while the energy and carbon cost of training is rarely reported. This study examines whether bin-wise quantization in CatBoost. reducing the feature bin count. can lower the computational and carbon cost of training such models without degrading predictive accuracy. Two public drug-review datasets are used: the UCI Drugs.com corpus (50,000 reviews, rating 1–10) and the WebMD corpus (50,000 reviews, satisfaction 1–5). Each review is represented by a 384-dimensional sentence embedding of its text, eight text-length and style statistics, a sentiment score, and two tabular features. CatBoost is used as the primary model because it produced markedly more stable generalization than the other libraries, and its bin count is directly exposed to the practitioner. Five CatBoost-based models are first compared per dataset under five-fold cross-validation: Ridge regression, standard CatBoost, and CatBoost with the feature bin count fixed at 64, 128, and 255. Hyperparameters are held constant across CatBoost variants to isolate the effect of the bin count. Accuracy equivalence is assessed with two one-sided tests (TOST) at a predefined margin of ±0.01 R², and training time and CO₂ emissions are tracked with CodeCarbon. The test R² values of the bin-reduced and standard CatBoost are statistically equivalent across both datasets and all three bin counts (TOST, all p < 0.001), while 64-bin reduction lowers training time and CO₂ emissions by around 48%. To test whether the effect is specific to CatBoost, LightGBM and XGBoost are evaluated under the same protocol; bin reduction lowers their training cost by roughly 25–60% while remaining statistically equivalent in accuracy (TOST, all p < 0.002), con-firming that the effect is general to histogram-based gradient boosting. SHAP analysis shows the review-text embedding is the dominant predictor, with sentiment and engagement as secondary signals. The work is presented as a sustainability-oriented empirical study: explicit control of the bin-count parameter exposes a favorable accuracy–efficiency trade-off applicable to biomedical review marketplaces.

An Explainable and Computationally Efficient Machine Learning Framework for Breast Cancer Diagnosis Using Predictive and Interpretability Analysis

Aziz Ur Rehman — 2026-09-01

Accurate breast cancer diagnosis is important for supporting early detection and improving clinical decision-making.. This study examines the theoretical and interpretative principles that support the machine learning algorithms applied to the Wisconsin Breast Cancer dataset and evaluates their ability to differentiate between benign and malignant neoplasms. To improve reliability, five-fold stratified cross-validation was additionally performed, and average performance values were considered. The logistic regression model achieved a high ROC-AUC value of 0.9960, whereas the performance of the random forest classifier was almost similar, with a slightly lower score. To enhance methodological transparency and interpretability, this study employs explainable AI techniques, including permutation importance, SHAP values, local and global explanation metrics, and partial dependence plots. These methods provide valuable insights into the extent to which tumour size and concavity contribute to malignancy prediction. A systematic study of computational complexity has been conducted, and the results show that the training and inference latency of logistic regression is significantly lower than that of random forests, indicating its potential suitability for future resource-constrained clinical decision-support investigations after external validation. The proposed framework achieved high discrimination performance while maintaining interpretability and low computational cost, suggesting its potential suitability for future clinical decision-support investigations after external validation.

AI Collaborative Framework for Gamified English Language Learning with Real-Time Encouraging Feedback

Arif Jawaid — 2026-08-14

Recent advances in transformer-based Natural Language Processing (NLP), speech intelligence, and adaptive machine learning have enabled the development of intelligent, real-time educational systems. This paper presents a scalable multimodal artificial intelligence (AI) framework for adaptive language learning that integrates text and speech processing within a unified architecture. Existing digital language learning platforms remain limited by delayed feedback, lack of multimodal interaction, and insufficient personalization. To overcome these challenges, the proposed framework combines transformer-based models (BERT and T5) for contextual text understanding and generation, DeepSpeech for automatic speech recognition, and predictive models, including XGBoost and Long Short-Term Memory (LSTM), for dynamic learner profiling and adaptive feedback generation. The system employs a unified multimodal pipeline to process textual and speech inputs, enabling low-latency, context-aware feedback on grammar, pronunciation, fluency, and vocabulary. A gamified adaptive learning mechanism dynamically adjusts task difficulty and reward progression based on real-time learner performance. Experimental evaluation on 120 tertiary-level learners over 16 weeks demonstrates significant improvements, including 38% in writing accuracy, 42% in speaking fluency, and 35% in engagement. Performance is validated using BLEU score, Word Error Rate (WER), F1-score, and latency metrics, confirming robustness and scalability. The proposed framework advances intelligent tutoring systems through integrated multimodal processing, real-time adaptability, and scalable AI-driven personalization.

Does Digital Tool Usage Translate Digital Competence into Workflow Efficiency? Explanatory and Predictive Evidence from AI-Augmented Banking

Meng Chai — 2026-09-01

Digital competence is frequently presented as a prerequisite for effective work with artificial-intelligence-enabled systems. Yet competence does not show that technology has been incorporated into task execution. This study examines whether task-embedded digital tool usage translates digital competence into perceived technology-enabled task efficiency. A cross-sectional questionnaire yielded 301 complete responses from a banking-sector sample. The proposed capability-enactment model was evaluated with reflective partial least squares structural equation modelling (PLS-SEM) and 5,000 bootstrap resamples. Its predictive implications were tested through repeated nested cross-validation. Ridge regression and random forests used competence-only, usage-only, and combined item-level feature sets. Digital competence was associated with digital tool usage (beta = 0.388) and perceived task efficiency (beta = 0.261). Tool usage retained an association with efficiency after competence was controlled (beta = 0.303). The indirect association was 0.118 (95% bootstrap CI 0.074-0.171), supporting partial statistical mediation. The combined random forest produced the lowest held-out RMSE (0.835) and highest pooled Q2predict (0.240). It also outperformed competence-only and usage-only forests in paired comparisons. Group permutation analysis assigned more predictive information to usage than to competence, although both were useful. The results identify digital tool usage as a proximal behavioural mechanism between capability and perceived workflow return. They also show why explanatory significance should be complemented by out-of-sample validation when designing or evaluating human-AI work systems.

Automated Detection and Segmentation of Pneumonia from Chest X-Ray Images Using a Convolutional Neural Network and U-Net

Mithoo Kailash — 2026-08-12

Pneumonia is a severe disease of the lungs with inflammation of the air sacs which can be a cause of severe sickness and end of life, particularly in children and elderly people. The most frequently used method to diagnose pneumonia is to use chest X-ray imaging, however manual diagnosis by radiologists may be timely and prone to errors. This paper presents a deep learning-based automated approach for lung segmentation and pneumonia diagnosis in chest X-ray (CXR) pictures. A Convolutional Neural Network (CNN) model was used to categorize the CXR pictures as either normal or pneumonia, and a U-Net model was used to segment the lung area. The models were trained and tested using a publicly available library of CXR pictures of both normal and pneumonia patients. Before being input into the model during training, each picture was shrunk to 224 × 224 pixels, normalized, and enhanced for optimal performance. It was compared with ResNet-50, MobileNet, Inception-V3, Vision Transformer (ViT), and the Swin Transformer models for the proposed CNN, and with a VGG U-Net model for the proposed U-Net. Compared to all the other comparison models, which varied from 90.43% to 93.34% accuracy and 0.82 to 0.88 recall value for pneumonia cases, the proposed CNN obtained an accuracy of 94.00% using a test set of 1150 CXR pictures and a 0.95 recall value for the pneumonia cases.

Creating an S-Box Using Chaotic Map for Improved Data Security: A Design Approach

Zarsha Nazim — 2026-09-01

In this era of digitization, data security is an inherent requirement over public internet. Due to advancement of Artificial Intelligence based attacks, data secuirty has become a challenge. For this purpose, substitution boxes (S-Box) with ciphers are used. Most of the existing S-Boxes use block ciphers that are currently created through chaotic maps. This article presents a dynamic novel technique of creating block ciphers that use innovative chaotic map approach to provide data security and integrity. This approach is initially designed and then tested against recently projected S-Box in fields of Nonlinearity (NL), Bijectiveness, Bit Independence Criterion (BIC), Strict Avalanche Criterion (SAC), Differential Approximation Probability (DP), and Linear Approximation Probability (LP). Time complexity analysis is also performed to evaluate the computational overhead of the proposed S-Box. The results of these Standard Evaluation Criteria signifies this approach to be particularly robust in the act of protection against attempts to cryptanalyze and proves its forte in data encryption and security.

AI Across Multiple Domains: A Comprehensive Review

Muhammad Ibrahim Tahir — 2026-08-02

Artificial Intelligence (AI) has gone from being a small branch of computation to becoming a ubiquitous and all-embracing influence in all aspects of human endeavour. This comprehensive The multidisciplinary review brings together information from more than 160 peer-reviewed publications from the previous 2020 to 2026, including healthcare and medical imaging, cybersecurity and intrusion detection, marketing, and others. and supply chain management, education, agriculture, consumer behaviour, finance and economics, In arts, humanities, law and the physical sciences. AI technologies are classified into eight main categories. It is compatible with families Deep Learning, Transformer Models, Explainable AI (XAI), Generative AI, Federated Learning. Know and be able to compare Quantum Machine Learning, Transfer Learning, and Optimisation Algorithms Deployment, performance and societal implications, both at the domain level and across domains. The deep findings are key points that are usually evident at a deeper level. learning dominates medical imaging with state-of-the-art segmentation accuracy; federated learning allows you to leverage the power of the group.federated learning lets you harness the power of the group. Leading the pack of reshaping technologies for cybersecurity is and hybrid CNN-LSTM architectures and for marketing is generative AI. Consumer engagement; and quantum-enhanced models are at the fore front of precision diagnostics andencryption. Wemap critical gaps that encompass cross-domain data interoperability, ethical considerations and ethics, and the representation of data. Critically discuss governance, explainability and equitable deployment, and present a single research agenda for the next decade.

Automated Skin Cancer Classification from Dermoscopic Images Using Fine-Tuned CNN Models

Muhammad Shoaib Rasheed — 2026-09-01

Skin cancer mostly melanoma cause a significant health risk as the dieses spread quickly and high risk of death if not detected early. The early detection of skin disease is essential but depending only on visual examination can be difficult for dermatologists and it may lead to delayed diagnoses. The deep learning based diagnostic systems can support clinicians by classifying skin lesions in a consistent manner and reducing variability in visual assessment. To automatically classify dermoscopic images we introduce a machine learning model that combines transfer learning with pre-trained CNNs. Three architectures VGG16, MobileNetV2 and EfficientNetB0 were evaluated on the MED-NODE dataset which contains images of melanoma and naevus that limited and imbalance. The methodology integrates the methods such as image pre-processing, extensive data augmentation, selective fine-tuning, hierarchical feature extraction and class weighting to improve model generalization and handle dataset limitations. The experimental results show that VGG16 achieved the highest performance with 82.5% accuracy, 84.67% weighted precision, 82.22% F1-score and 91.25% AUC. MobileNetV2 achieved moderate performance, whereas EfficientNetB0 performed poorly. This highlights the importance of deeper pretrained features and careful adaptation for small and imbalanced medical datasets. Analysis of per class performance, confusion matrices and training/validation curves shows that VGG16 extracts low-level features including edges, textures and also identifies higher-level patterns like asymmetry, borders and lesion shape which helps distinguish melanoma from naevus more reliably. The results show that transfer learning with deep CNN models improves automated skin lesion classification. The performance further improves with data pre-processing, data augmentation, class weighting and fine-tuning strategies. We proposed a machine learning framework for melanoma detection that helps clinicians to achieve more reliable diagnostic results in clinical practice. The framework can be evaluated on larger datasets using hybrid models to improve both performance and interpretability.

Towards Trustworthy Software-Defined Network Security: An Explainable and Computationally Efficient Machine Learning Framework for Intrusion Detection

Suhail Ashfaq Butt — 2026-09-01

Software-defined networking (SDN) offers flexibility and scalability but also introduces new cybersecurity challenges. We propose a scalable and transparent intrusion detection system (IDS) for SDNs using machine learning models that balance accuracy and computational efficiency. Our results show that the decision tree model achieves 99.6% accuracy with minimal missed attacks and lower computational costs compared with random forests. The training time increased with the dataset size, but the prediction time remained stable, even with 24,000 samples. Our simulation results showed that the training complexity of Random Forest increases with the number of samples, whereas the calculation time for inference is comparatively low–a finding that makes Random Forest particularly suitable for real-time detections in Software-Defined Networking (SDN) environments. By using the SHAP-based explainability analysis, we explain the most important traffic characteristics, such as the length of the packets and the destination port, which are used to decide the decision process of the model. This methodological approach ensures high recall detection without compromising performance, thus providing a high-performance and efficient solution for SDN security.