IJACSA Vol. 17 Issue 8 (2026)

1

An Intelligent Ensemble Learning Framework for Visibility and CAVOK Prediction to Support Logistics Operations

Author 1: Omar Alghushairy Author 2: Raed Alsini Author 3: Ahmed Alamri Author 4: Saud Yonbawi Author 5: Ayman Yafoz Author 6: Xiaogang Ma

The supply chain operation is fully dependent on weather conditions due to the damage that produce may face during bad weather or accidents that may occur because of cloud cover, especially in Saudi Arabia, which experiences frequent cloud cover throughout the year. In this study, a new weather prediction model based on machine learning and optimization algorithms is proposed to forecast visibility distance and Cloud and Visibility OK (CAVOK), a term indicating good weather, while also estimating cloud cover, which affects flight operations. The proposed system follows two main steps: feature selection and prediction, to predict CAVOK and visibility distance. In the feature selection phase, Fick’s Law Algorithm (FLA) is used to choose the optimal parameters that can serve as indicators in the prediction phase, enhancing the proposed system's performance. In the prediction phase, a new two-stage ensemble algorithm is used to predict sky CAVOK and visibility distance. In the first stage, a new voting ensemble algorithm consisting of Bagging Regressor and Extra Trees Regressor algorithms is proposed to predict the visibility distance. The predicted visibility distance value is then used alone to predict sky CAVOK using the CatBoost Classifier in the second stage. The experimental results show that the proposed system demonstrates reasonable predictive performance, obtaining 529.7506 m, 1174.6118, and 0.3233 for Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R², respectively, in predicting visibility distance. Additionally, the system achieved competitive results while maintaining robust predictive capability in predicting sky CAVOK.

Weather forecasting sky CAVOK extra tree supply chain visibility distance Saudi Arabia

Abstract doi.org/10.14569/IJACSA.2026.0170801 PDF

2

AI-Mediated STEM Problem Solving: Activity Distribution and Functional Roles in a Systematic Review

Author 1: Mika Lim

Artificial intelligence (AI) is increasingly embedded in STEM problem-solving activities, yet existing reviews have generally organized the literature by technology type or learning outcome. This review shifts the analytic focus from what AI can do to how it is integrated into students' STEM problem solving. It examined where AI was positioned within problem-solving activity, what functional role it performed, and how its functions or outputs were connected to students' subsequent judgments and actions. Following PRISMA 2020, searches of Scopus, Web of Science, ERIC, and backward references identified 29 English-language journal articles published from 2017 through June 4, 2026. Study-level coding mapped AI use across seven non-linear activity categories, while case-level analysis examined 37 reporting-eligible AI–student interactions in which the AI function or output, activity context, and subsequent student action were identifiable. AI was reported most often in two activity categories: Problem Understanding and Exploration, and Revision and Improvement. Four recurring functional roles were identified: Conceptual Resource, Generative Resource, Operational Component, and Adaptive Scaffold. The same technology assumed different roles depending on the activity category and how students used its output. The principal contribution of this review is to shift attention from AI's technical capabilities and associated learning outcomes to the interaction between AI outputs and students' subsequent activity. By linking each AI function or output to a problem-solving activity category and a subsequent student action, the review shows that AI's role is defined by how students use its output in context, not by the technology itself. The findings are descriptive of the selected corpus and its reporting practices; they should not be interpreted as prevalence estimates or evidence of causal learning effects.

Artificial intelligence STEM education AI-mediated learning human-AI interaction functional roles systematic review

Abstract doi.org/10.14569/IJACSA.2026.0170802 PDF

3

Five Whys as an Epistemic-Honesty Scaffold for Multi-Agent LLM Analysis of Industrial Time Series

Author 1: Yoshihiro Ochi Author 2: Yasunobu Uchiyama

Multi-agent large language model (LLM) pipelines increasingly analyse industrial time series. Where the true causes lie outside the recorded signal, they tend to report confident, plausible-sounding mechanisms not grounded in the data — a false root-cause handed to an engineer. We port Toyota’s Five Whys onto such a multi-agent system (MAS) not as a decomposition tool but as a reviewer frame, and find it functions as an epistemic-honesty scaffold. On a non-public power-consumption series from a coating application process, whose true drivers are exogenous to the data, a single Five Whys chain drives the MAS to surface three grounded-sounding mechanisms and, by repeatedly asking where in the data is this grounded?, retract each in turn, converging on the honest verdict: the cause is exogenous to, and unverifiable from, this signal. A matched baseline, run to the same depth without the reviewer, commits to one such mechanism as a confident but unverified cause. In the three-seed mean —a directional, mean-level effect with large per-seed variance and occasional sign reversals — the scaffold raises contestation on the hard series, whereas on a clean benchmark with an in-data root it instead reconciles competing narratives onto that root, adapting to whether a groundable root exists rather than adding noise. The honest verdict is the right target: the strongest model states it unaided, so the scaffold’s value concentrates where the model cannot self-regulate. We frame it as a discipline for human–AI coexistence: it lets a practitioner correctly receive what the MAS does and does not know.

Five Whys multi-agent systems large language models root-cause analysis epistemic honesty causal abstention industrial time series argumentation framework human–AI collaboration

Abstract doi.org/10.14569/IJACSA.2026.0170803 PDF

4

A Reproducible Mobile-Cloud Framework for Preliminary Non-Invasive Anemia Screening from Palpebral Conjunctiva Images

Author 1: Enrique Lee Huamaní Author 2: Erasmo Montufar-Barrientos Author 3: Victor Romero-Alva Author 4: Linett Velasquez-Jimenez

Non-invasive anemia screening from palpebral conjunctiva images may support preliminary triage when access to laboratory hemoglobin testing is limited. This study presents and evaluates a reproducible mobile-cloud framework that integrates guided image capture, automated eye-region localization, region-of-interest extraction, color normalization, image-quality control, deep-learning inference, probability calibration, longitudinal storage, and user feedback. The evaluation used five-fold image-level cross-validation, a locked hold-out image set, preprocessing ablation, mobile efficiency benchmarking, and a formative us-ability assessment. Among the evaluated backbones, EfficientNet-B0 achieved a cross-validated accuracy of 87.84%±0.94, sensitivity of 89.20%±1.38, specificity of 86.78%±0.72, F1-score of 86.68%±0.77, and area under the receiver operating characteristic curve of 0.927±0.007. On the locked hold-out set of 120 images, the model achieved 85.83% accuracy, 87.93% sensitivity, 83.87%specificity, 83.61% precision, 85.71% F1-score, and 0.952 AUC. The ablation analysis showed incremental gains from automated ROI extraction, white balancing, contrast normalization, and image-quality rejection. The results indicate that predictive performance, calibration, acquisition quality, and deployment cost should be evaluated together when designing mobile screening support systems. The tool is intended for preliminary screening and does not replace laboratory confirmation.

Anemia screening conjunctival image analysis mobile health deep learning EfficientNet mobile-cloud computing reproducible research

Abstract doi.org/10.14569/IJACSA.2026.0170804 PDF

5

Reversible Visual Obfuscation via Embedded Recovery Metadata: Combining Median‑Filter Irreversibility with Cryptographic Restoration

Author 1: Andranik Karakhanyan

Although cloud storage platforms are widely used to safeguard personal images, data leakage and unauthorized access remain persistent threats, exposing sensitive visual content. Linear obfuscation methods such as Gaussian blur can be reversed by deconvolution when kernel parameters are known, while non-linear median filtering permanently destroys details but also eliminates any possibility of legitimate recovery. This study introduces a reversible obfuscation framework that combines the practical irreversibility of large-radius median filtering with an embedded, encrypted restoration payload. A sensitive image region is first blurred with a non-invertible median filter, making it resistant to both deconvolution and AI‑based reconstruction. Simultaneously, a losslessly compressed and AES‑256‑GCM‑encrypted copy of the original region is hidden in the least significant bits of the blurred block. Unauthorised viewers—including any party that intercepts the image through a cloud leak—see only the irreversibly obfuscated content; an authorised user with the correct passphrase can decrypt and restore the original region with pixel‑perfect fidelity. Experimental results on facial and document images demonstrate that the obfuscated region withstands state‑of‑the‑art AI restoration while the embedded payload enables lossless recovery in under 50 ms, with an average overhead of only 0.41 bits per pixel (using a single embedding channel) and no visible artefacts. The method offers a practical balance between strong, irreversible privacy and cryptographically controlled recoverability, making it suitable for secure archiving, forensic analysis, and privacy‑preserving communications.

Sensitive information visual privacy image processing blur filter median filtering data obfuscation reversible concealment

Abstract doi.org/10.14569/IJACSA.2026.0170805 PDF

6

Automatic Generation and Optimization Algorithm for Brand Visual Identity System Based on Multimodal Generative Adversarial Network

Author 1: Yunfei Gong Author 2: Xiaoyue Cao

A Brand Visual Identity System (BVIS) is a crucial carrier for conveying a brand's core values, but traditional manual design struggles to meet the demands of efficient and personalized design in the digital age. Addressing the shortcomings of existing Multimodal Generative Adversarial Networks (MM-GANs) in BVIS design, such as semantic-visual mapping bias, fragmented element styles, and a lack of quantified iteration mechanisms, this study integrates attention mechanisms, perceptual loss, style consistency constraints, and adaptive optimization strategies to construct an improved automatic BVIS generation and optimization framework. The model utilizes publicly available datasets for preprocessing and alignment of text, color, and graphic multimodal data. Text semantic encoding is achieved through Bidirectional Encoder Representations from Transformers (BERT), and multimodal feature fusion is accomplished using an attention matrix. Adversarial training is conducted using an improved U-Net and PatchGAN. The model is then compared and tested with Generative Adversarial Network (GAN) series models, diffusion models, and Contrastive Language-Image Pre-Training (CLIP) guided generative models. Experiments show that the model outperforms the control model in metrics such as Fréchet Inception Distance (FID), Structural Similarity Index Measure (SSIM), and Semantic Matching (SM), demonstrating excellent robustness and generalization ability in scenarios with noise interference, cross-industry collaboration, and modality loss. Ablation experiments prove that multimodal feature alignment is key to improving model performance. The generated visual elements exhibit a consistent style and good semantic matching. This study presents a practical intelligent brand design solution and analyzes the ethical risks of copyright infringement, counterfeiting, and misuse of AI-generated logos, providing insights for compliant technology applications and expanding the engineering application of multimodal generation models in brand visual design.

Brand digitalization visual identity system multimodal generative adversarial network attention mechanism

Abstract doi.org/10.14569/IJACSA.2026.0170806 PDF

7

Integrated Health Belief and User Engagement Model for Physical Activity Apps

Author 1: Nur Farahin Mohd Johari Author 2: Nazlena Mohamad Ali Author 3: Mohamad Hidir Mhd Salim

Physical activity applications support monitoring, motivating, and sustaining healthy behavior, yet long-term engagement remains challenging. This study validates an integrated Health Belief Model and User Engagement Model for explaining behavior change among physical activity application users. Survey data from 423 users were analyzed using Partial Least Squares Structural Equation Modeling, followed by a four-week Strava study involving 17 participants and post-usage interviews. The results showed that Healthy Behavior Adoption and Health Technology Engagement significantly influenced behavior change, with technology engagement exerting the stronger effect. Perceived usability and reward factors were the strongest predictors of technology engagement, while self-efficacy and objective barriers were the strongest predictors of healthy behavior adoption. Focused attention and cues to action were not significant. Real-world usage and interview findings supported the model by showing that usability, meaningful feedback, and perceived progress were more influential than continuous attention or repeated reminders in supporting sustained physical activity behavior.

Mobile health physical activity applications health belief model user engagement model behavior change technology engagement

Abstract doi.org/10.14569/IJACSA.2026.0170807 PDF

8

Evaluating Government Website Readiness for Generative AI Search Engines: A Framework and Empirical Assessment

Author 1: Hussein Ali Bahadi

The rapid adoption of generative artificial intelligence (GenAI) search engines is changing how citizens obtain government information. Unlike conventional search engines, which return ranked links, GenAI systems synthesise answers directly from retrieved sources, positioning government websites as inputs to AI-mediated information ecosystems rather than as destinations. To our knowledge, no systematic framework currently exists for determining whether government websites are prepared for this form of access. This study develops and validates an AI Readiness Index for government websites through a three-phase sequential mixed-methods design. The framework is validated perceptually through a one-round Delphi-informed expert survey (n = 18) and a citizen survey (n = 85) analysed using partial least squares structural equation modelling. The framework is applied to government portals across 14 jurisdictions using a documented scoring protocol. The expert panel confirmed content validity for Data Foundation, Technical Optimisation, and Content Quality; Discoverability met the consensus threshold marginally, and Governance did not reach it (I-CVI = 0.67). The structural model identifies Technical Optimisation and Content Quality as significant predictors of perceived AI readiness, explaining 55 per cent of the variance, while Discoverability and Governance are non-significant. Portal assessment reveals wide disparities, with relative strength in data foundations and technical optimisation and consistent weakness in governance, ethics, and accountability. Because the index measures structural properties that condition AI retrieval rather than testing retrieval performance directly, its scores are proxy indicators of readiness rather than measures of realised AI output quality.

Generative AI government websites AI readiness E-government retrieval-augmented generation SmartPLS digital transformation evaluation framework

Abstract doi.org/10.14569/IJACSA.2026.0170808 PDF

9

A Hybrid Econometric and Machine Learning Framework for Identifying Nonlinear Dynamics Between Digital Economy and Carbon Emissions

Author 1: Jinghan Li

The rapid expansion of the digital economy has reshaped global production systems and energy consumption patterns, yet its environmental consequences remain theoretically ambiguous. This study investigates the nonlinear relationship between digital economy development and carbon emissions using a cross-country panel dataset covering 45 countries during 2012–2024. A hybrid econometric and machine learning framework is developed to identify the structural dynamics underlying the digital economy–carbon emissions nexus. The empirical analysis combines fixed-effects estimation, System Generalized Method of Moments, instrumental variable estimation, and threshold regression with Gradient Boosting Decision Trees and SHAP-based interpretation. The results reveal a robust inverted-U relationship between digital economy development and carbon emissions, with the quadratic specification yielding a turning point of approximately 0.71 and the panel threshold model indicating a significant threshold value of 0.742. This indicates that digital transformation initially increases emissions through infrastructure expansion and energy demand, but subsequently contributes to emission mitigation through technological progress, energy efficiency improvement, and industrial restructuring. The nonlinear transition pattern is consistently validated across multiple identification strategies, with machine learning evidence further confirming the heterogeneous marginal effects of digital development across different development stages. Mechanism analysis shows that energy efficiency enhancement, technological innovation, and industrial upgrading are key channels through which digitalization promotes low-carbon transformation. These findings highlight that the environmental impacts of the digital economy are not inherently positive or negative but evolve dynamically with technological maturity and economic structure. Policy strategies should therefore emphasize the quality-oriented development of digital infrastructure and strengthen complementary innovation capabilities to maximize the emission-reduction potential of digital transformation.

Digital economy carbon emissions nonlinear relationship system GMM machine learning energy efficiency industrial upgrading

Abstract doi.org/10.14569/IJACSA.2026.0170809 PDF

10

A Multi-Layer Hyper-Personalization Pipeline for Predicting Next User Actions Across Web, App, Email, and Social Channels

Author 1: Kohei Arai

Hyper-personalization systems that unify customer behavior across Web, App, Email, and social channels are increasingly central to digital business strategy, yet published evaluations rarely isolate which modeling component—collaborative filtering, sequential prediction, or behavioral clustering—actually drives predictive value, and rarely validate a full pipeline end-to-end before committing to production data collection. This study presents a complete hyper-personalization pipeline spanning unified behavior aggregation, recency/frequency/content feature engineering, item-based collaborative filtering, sequential next-action modeling (first-order Markov chains and LSTM networks), and K-means behavioral clustering, validated on a simulated multi-channel dataset (80,997 events, 600 users, 90 days) constructed with five known latent personas so that recovered structure could be checked against ground truth. The LSTM next-action model reached 34.5% validation accuracy against a 10.0% uniform-chance baseline and a 16.0% majority-class baseline, and outperformed the order-1 Markov chain by approximately 6 percentage points once evaluated on a matched held-out split. An ablation study further shows that 1) LSTM accuracy is insensitive to capacity across a 50-fold parameter range, suggesting that the bottleneck is behavioral signal rather than model size; 2) collaborative-filtering hit-rate@2 is strong despite weak absolute cosine similarities, indicating that relative ranking survives even when category-level content granularity limits similarity magnitude; and 3) clustering quality is substantially higher on a reduced recency/frequency/channel-mix feature set than on the full feature set including content preferences, implying that content preferences are better used downstream in recommendation than as clustering inputs. I report these findings, together with the pipeline’s limitations on simulated versus real data, as a feasibility baseline for practitioners designing multi-channel personalization systems.

Hyper-personalization collaborative filtering sequential recommendation LSTM Markov chain customer segmentation multi-channel analytics ablation study

Abstract doi.org/10.14569/IJACSA.2026.0170810 PDF

11

Deep Learning Approaches for Apple Disease Classification and Detection: A PRISMA 2020 Systematic Literature Review

Author 1: Brahim Ouben Hssain Author 2: Khalil Ladrham Author 3: Noureddine El Barbri Author 4: Rachid El Ayachi

Apple diseases cause great losses in fruit yield and quality, while manual diagnosis is time-consuming, subjective, and difficult to scale across orchards. Deep learning has thus become a staple in image-based disease recognition, but evidence remains scattered across classification and object-detection paradigms, heterogeneous datasets, and inconsistent evaluation protocols. This systematic literature review integrates 30 peer-reviewed studies published between 2020 and 2026, selected from 180 database records using a PRISMA 2020-aligned protocol. The final corpus includes 22 image-classification studies and eight object-detection studies. One non-peer-reviewed preprint was kept as contextual evidence only and was excluded from all corpus counts and comparative analyses. Architectures, datasets, preprocessing and augmentation strategies, training configurations, evaluation metrics, evidence of generalization, and deployment characteristics are discussed separately for the two task paradigms. Reported classification accuracies vary from 91.0% to 99.99%, whereas detection studies report mAP@0.5 values from 82.1% to 99.99%; these ranges are descriptive and are not pooled estimates. These values are heavily dependent on dataset composition and evaluation design: controlled-background datasets often yield near-perfect scores, while field transfer can lead to a drop of almost 30 percentage points. CNNs still dominate the field, but hybrid transformer-based attention mechanisms, multi-scale feature fusion, class-imbalance-aware training, and lightweight YOLO variants are gaining traction. Long-standing limitations are the lack of reporting of hyperparameters, an over-reliance on accuracy, the lack of external validation, heterogeneity in annotation, the lack of uncertainty analysis, and limited reporting of latency, memory, and energy consumption. Accordingly, the research agenda emphasizes standardized real-orchard benchmarks, cross-dataset validation, calibrated and explainable predictions, reproducible experimental protocols, and deployment-aware model design.

Apple disease recognition deep learning convolutional neural networks object detection PRISMA 2020 systematic literature review precision agriculture

Abstract doi.org/10.14569/IJACSA.2026.0170811 PDF

12

When Calibrated Detectors Meet New Attacks: Per-Category Reliability of Machine-Learning Intrusion Detection Under Distribution Shift

Author 1: Khalid Alalawi

Machine-learning intrusion detectors are usually reported with accuracy or F1 on a single train and test split, and their confidence scores are often read operationally as probabilities without an explicit calibration check. We test that assumption. We measure the reliability of the predicted probabilities of three classifiers, logistic regression, random forest, and XGBoost, on two benchmarks, NSL-KDD and UNSW-NB15, in two settings: when training and test data share a distribution, and under the shift each benchmark's official split already contains. In distribution, all three detectors are well calibrated, with top-label expected calibration error at most 0.03. Under shift, the outcome depends on its content. On NSL-KDD, whose test set contains attack types absent from training, top-label calibration error rises from at most 0.007 in distribution to between 0.170 and 0.209, and the error is largest on the benign and Remote-to-Local classes, so unfamiliar attacks are labeled normal with high confidence. On UNSW-NB15, whose split keeps the same attack categories, the error stays far smaller, between 0.025 and 0.068, even though a domain classifier detects a clear shift on both benchmarks. Recalibrating on training-distribution data does not reliably transfer to the shifted test set. Supervised target-domain recalibration on a held-out labeled sample from the shifted distribution substantially reduces the calibration error and recovers much of the minority-class recall on NSL-KDD. Most of the aggregate top-label ECE reduction is obtained with about one hundred labeled target samples. This remedy requires labeled observations from the shifted environment and therefore describes reliability after target labels become available, not reliability against attacks that remain entirely unseen. A decision-curve analysis and expected-cost comparison show that an uncalibrated score used at its nominal cost threshold is very costly, and that both recalibration and a target-tuned threshold reduce operating cost, with recalibration additionally improving the multiclass probabilities. A leave-one-attack-out study reproduces the failure and indicates that the attack family for which performance degrades substantially is the one least separable from benign traffic, not the one most distinct from the other attacks.

Intrusion detection probability calibration distribution shift expected calibration error decision curve analysis class imbalance network security

Abstract doi.org/10.14569/IJACSA.2026.0170812 PDF

13

Explainable Machine Learning for Cardiovascular Disease Prediction Using BRFSS Health Indicators

Author 1: Sakchai Tangprasert Author 2: Suwit Chantasen Author 3: Nalinpat Bhumpenpein Author 4: Yuenyong Nilsiam Author 5: Siranee Nuchitprasitchai

Cardiovascular disease remains a major public health burden and is associated with demographic, behavioral, and chronic-health characteristics. This study evaluates explainable machine learning for cross-sectional classification of self-reported cardiovascular disease status using the 2020 CDC Behavioral Risk Factor Surveillance System (BRFSS)-derived Personal Key Indicators of Heart Disease dataset. After removing 18,078 exact duplicate feature-target rows, 301,717 records and 17 predictor variables were retained. The target variable, HeartDisease (Yes/No), indicates whether a respondent had ever been diagnosed with cardiovascular disease or had experienced a heart attack; therefore, the task is classification of prevalent disease status rather than longitudinal prediction of future disease incidence. Six machine learning families were compared using stratified train-test splitting, 5-fold cross-validation, cost-sensitive learning, and hyperparameter tuning. The tuned XGBoost model achieved the highest test-set F1-score of 0.3939 (approximate 95% CI: 0.384–0.403) and a ROC-AUC of 0.8362 (approximate 95% CI: 0.829–0.843). SHAP analysis showed that age category, self-rated general health, sex, smoking status, and body mass index contributed most strongly to the model outputs. These SHAP values describe model dependence and should not be interpreted as causal effects or as novel epidemiological risk-factor discoveries. A prototype business intelligence dashboard was developed for research-oriented communication of model performance and feature contributions. The results support transparent classification of self-reported cardiovascular disease status for analytical and preliminary screening contexts, not future-risk forecasting or direct clinical diagnosis.

Cardiovascular disease machine learning explainable artificial intelligence cardiovascular disease status classification behavioral risk factor surveillance system cross-industry standard process for data mining (CRISP DM)

Abstract doi.org/10.14569/IJACSA.2026.0170813 PDF

14

Transformer Health Index Prediction Using Static and Temporal Learning Models

Author 1: Omar Ahmed Mohammed Hamood Al-Shaikh Author 2: Arfah Ahmad Author 3: Ahmad Jazlan Haja Mohideen Author 4: Muhammad Sharil Yahaya

Transformer health index (THI) prediction supports condition-based maintenance by mapping dissolved-gas, oil-quality, and furan indicators to an interpretable asset-condition score. This study develops a supervised benchmarking framework using 3,392 transformer oil diagnostic records from 510 transform-ers. Thirteen diagnostic features from dissolved gas analysis, oil quality analysis, and furan analysis are evaluated across three modelling groups: static machine learning, static deep learning, and variable-length temporal deep learning. Transformer-level splitting is applied to reduce data leakage from repeated trans-former histories. Static gradient-boosted tree models achieved the strongest results. XGBoost obtained the lowest test root mean square error of 3.4254 with a coefficient of determination of 0.9780 and health-index class accuracy of 88.20 percent. The best temporal model, the liquid time-constant neural network (LTC-LNN), achieved a test root mean square error of 4.8691 and a coefficient of determination of 0.9567. Permutation feature importance identified 2FAL, C2H2, and dielectric break-down as the most influential predictors. For the present dataset, static gradient-boosted tree models produced the strongest ob-served point-estimate performance, while LTC-LNN remained a promising temporal-learning alternative for repeated diagnostic histories. However, the uncertainty analysis indicates that small numerical differences among the leading static models should not be interpreted as definitive evidence of model superiority.

Transformer health index machine learning deep learning liq-uid neural networks oil diagnostic data condition-based mainte-nance smart grid asset management

Abstract doi.org/10.14569/IJACSA.2026.0170814 PDF

15

ROMBandAR: Development of an AR-Based Wrist ROM Assessment Prototype for Rehabilitation

Author 1: Nurul Aimi Johan Author 2: Wan Rizhan Author 3: Normala Rahim

Wrist range of motion (ROM) assessment is important in rehabilitation because wrist movement supports hand positioning and functional hand use. Limited wrist movement can be caused by a variety of conditions, including neurological conditions such as stroke and musculoskeletal or orthopedic injuries such as distal radius fractures or tendon repairs. In current practice, wrist ROM is commonly assessed using a goniometer. Although the tool is widely used, the reading still depends on correct positioning, therapist handling, and visual judgment. Augmented reality has the potential to support this process because it can show digital feedback together with the real wrist movement. However, AR is still not widely explored for wrist-specific ROM assessment compared with virtual reality, robotics, and wearable or sensor-based systems. This study presents the development of ROMBandAR, an AR-based wrist ROM assessment prototype for rehabilitation. The prototype uses a wrist armband as a model target for AR tracking. The armband was first scanned using KIRI Engine software, then trained using Vuforia Model Target Generator, and lastly imported into Unity for AR integration. MediaPipeUnityPlugin was also integrated to show hand landmarks on the detected hand via the smartphone camera. After that, custom C# scripts were developed to display wrist ROM readings, classify movement, show real-time feedback, save assessment readings, and support report submission for therapist review. This study follows the Design Science Research Methodology and, at this stage, focuses on development-level testing. Three test cases were carried out: device compatibility testing, application functionality testing, and screenshot-based ROM reading consistency testing. The results show that ROMBandAR can run on different Android devices, successfully detect the wrist armband, display hand landmarks, show ROM readings, and save assessment results. These findings provide an early foundation for future evaluation of the proposed framework, especially the extrinsic feedback provided during wrist ROM assessment. Repeated trials, inter-rater reliability, direct goniometer measurement of the wrist, and further evaluation of the feedback were not part of this development stage and are planned as future work.

ROMBandAR augmented reality wrist range of motion rehabilitation pose estimation MediaPipe vuforia extrinsic feedback

Abstract doi.org/10.14569/IJACSA.2026.0170815 PDF

16

Intelligent Booking Systems for Natural Resources and Urban Development: A Geospatial SLR

Author 1: Stephanie Anak Nanta Author 2: Jack Febrian Rusdi Author 3: Louisa Suly Anak Nanta Author 4: Imtiaz Ali Brohi Author 5: Vicente Aquino Pitogo

Rapid urbanization and increasing pressure on natural resources demand efficient, transparent, and sustainable management systems. However, existing booking systems in government agencies often operate in silos and lack integration with geospatial data. This study conducts a Systematic Literature Review (SLR) to identify essential components of intelligent booking systems (IBS) and research gaps in the context of Natural Resources and Urban Development with geospatial data integration. The SLR follows a three-phase protocol (Planning, Execution, Reporting). The literature was searched in IEEE Xplore, SpringerLink, ScienceDirect, and ACM Digital Library for 2021–2025. From an initial pool of 40 articles, 16 were selected for in-depth analysis after eligibility and quality assessment (scoring ≥ 3 out of 5). The analysis reveals six main component categories, with Resource Management (n=12) and Booking Management (n=11) as the foundational layers, whereas Geospatial Integration was addressed in only 4 of the 16 studies. Critically, no existing system integrates Sentinel satellite data (Sentinel-1/2), and no optimisation model combines Multi-Criteria Decision Making (MCDM) with GIS for this specific domain. This SLR provides a comprehensive component framework and systematically maps four key research gaps, laying a foundation for developing geospatially-integrated intelligent booking systems for smarter natural resource and urban facility allocation.

Intelligent booking systems natural resource management Geographic Information Systems (GIS) satellite remote sensing data mining Multi-Criteria Decision Making (MCDM)

Abstract doi.org/10.14569/IJACSA.2026.0170816 PDF

17

A Smart Transformer-Based Model to Classify Extremist Behavior on Social Media

Author 1: Tamer Salah Author 2: Hazem M. El-Bakry Author 3: Amira Rezk

In recent years, social media has been used as a fertile environment for spreading radical ideologies through extremist organizations and groups to recruit new individuals capable of implementing their schemes. This has led to an enormous amount of extremist data that can be used by research agencies to analyze and understand the behavioral and linguistic characteristics associated with extremist language, thereby contributing to the development of policies and laws that help to protect societies from this threat. This study proposes a smart transformer-based model for collecting and analyzing data on extremism from social media to be classified based on whether or not it relates to extremism. The proposed framework begins by collecting data from Twitter by using keywords related to extremism and then filtering and reprocessing the data to implement the word embedding process that relies on a bi-directional LSTM network to provide the best representation vector for each word in the input sentence that can provide the true meaning of each word based on its position in the input sentence. Here comes the importance of extracting linguistic and semantic features of the text by building a matrix of features that can be extracted from the text, divided into three categories related to surface features, polar features, and specific domain features. Finally, after effectively preparing data and extracting features, the vector representing each sentence is inserted into the proposed smart transformer-based model to carry out the classification task, as the development steps focus on increasing the model's ability to raise the standard weight of words related to extremism in the inserted sentence, thereby contributing to increasing the accuracy of the model to classify the text. The results demonstrated the ability of the proposed model to achieve a high accuracy rate on the F1-score benchmark of 91.6% as well as 96.7% on the accuracy standard. The proposed model was able to outperform the accuracy of 5 models of other algorithms that achieved high results in the classification task, exceeding the closest competition of SVM-Linear by 0.7% at the F1-score benchmark and 0.5% at the overall rating accuracy level.

Social media transformer-based model word embedding Bi-LSTM network additive attention

Abstract doi.org/10.14569/IJACSA.2026.0170817 PDF

18

Benchmarking Bio-Inspired Metaheuristics for Load-Aware Task Offloading in Hierarchical IoT-Fog-Cloud Ecosystems

Author 1: Lamia Oualili Author 2: Mohamed EL Ghmary Author 3: Hassan Echoukairi

The multi-objective nature of task scheduling for novel latency-sensitive applications in IoT-Fog-Cloud hierarchies presents persistent challenges, where optimizing one QoS metric often degrades another. Although bio-inspired algorithms provide adaptive solutions, existing comparative studies are constrained by narrow algorithmic scopes, limited evaluation metrics, and a notable absence of statistical validation. To bridge this gap, we introduce a novel unified benchmarking framework that systematically evaluates five prominent metaheuristics — Particle Swarm Optimization (PSO), Genetic Algorithm (GA), Bacterial Foraging Optimization (BFO), Ant Colony Optimization (ACO), and Artificial Bee Colony (ABC) — for load balancing within a three-tier architecture using the iFogSim simulator. We evaluate performance across three QoS dimensions (response time, makespan, and load imbalance degree) under three escalating workload intensities on heterogeneous infrastructure, with all experiments repeated over 10 independent runs. Statistical significance is assessed via the Wilcoxon signed-rank test against GA, selected as a widely established and computationally stable baseline. Our findings reveal no universally optimal algorithm. GA consistently maintains mean response times below 30.5 ms across all workloads, while PSO and BFO remain statistically indistinguishable from GA on makespan, highlighting their interchangeability under specific conditions. Conversely, ABC uniquely excels in distribution equity, reducing load imbalance from 2.64% to 1.32% as task density increases. ACO, however, incurs statistically significant penalties across all metrics, suffering from a pronounced cloud-bias that elevates load imbalance to 11.0% —up to eight times higher than its counterparts. Collectively, these results confirm the inherent trade-offs between latency, efficiency, and fairness. This reference benchmark delivers a reproducible, statistically validated performance baseline, offering system architects a clear empirical foundation for adaptive algorithm selection in heterogeneous Fog-Cloud environments.

Load balancing IoT-Fog-cloud metaheuristic algorithms iFogSim QoS optimization wilcoxon signed-rank test task scheduling edge computing

Abstract doi.org/10.14569/IJACSA.2026.0170818 PDF

19

Emerging Technologies and Cybersecurity in Critical Information Infrastructure and Industrial Control Systems: An Integrated Cyber Risk Pathway Model

Author 1: Jennita Rao Appanah Appayya Author 2: Sandhya Armoogum Author 3: Kaleem Ahmed Usmani

The digital transformation of Critical Information Infrastructure (CII) and Industrial Control Systems (ICS) through Industry 4.0 technologies introduces significant cybersecurity challenges. While existing research examines technologies individually, little attention has been given to how their combined adoption reshapes the overall threat landscape. This study presents a Multivocal Literature Review, synthesising evidence from 41 academic and industry sources (January 2010–June 2026) and proposes an Integrated Cyber Risk Pathway Model that traces how technology adoption introduces interconnected vulnerabilities, expands threat actor capabilities, produces cyber-physical impacts, and ultimately defines resilience requirements. Three findings emerge: 1) emerging technologies play a dual role, enhancing operational capability while expanding the attack surface; 2) cyber-attacks have evolved from specialist ICS operations to ecosystem-level compromises exploiting supply chains and shared platforms; 3) the resulting vulnerabilities are systemically interconnected, creating risks that prevention-focused cybersecurity alone cannot fully address. The study argues that protecting modern critical infrastructure requires a shift to resilience-centred strategies supported by governance, secure system design and cross-sector collaboration.

Critical Information Infrastructure (CII) Industrial Control Systems (ICS) emerging technologies industry 4.0 IT/OT convergence cyber resilience cyber-physical systems multivocal literature review integrated cyber risk pathway model

Abstract doi.org/10.14569/IJACSA.2026.0170819 PDF

20

MalBERT-Temporal: Transformer-Based Zero-Day Malware Detection in Windows Executable Binaries Under Strict Temporal Isolation

Author 1: Manar Alanazi Author 2: Israa Alsiyat

In current malware detection benchmarks, random train/test splits are commonly used, allowing temporal leakage to occur and obscuring performance degradation caused by concept drift. Moreover, traditional classifiers that operate on a one-dimensional PE feature vector do not explicitly model long-range interactions between structurally distant feature groups. This study introduces a Transformer-based malware detection approach named MalBERT-Temporal that reshapes the 2,381-dimensional BODMAS PE feature vector into 16 contiguous feature-group tokens and then processes these tokens with Transformer encoder layers employing multi-head self-attention to model interactions among all tokens. The proposed approach is evaluated using a strict temporal protocol in which the training data comprise only pre-2020 samples, and the test data span the complete 2020 evaluation timeline. Each model is independently calibrated using a fixed validation set with a false-positive-rate budget of 0.1% and is evaluated monthly and on malware families unseen during training. At the calibrated security threshold, MalBERT-Temporal achieves an F1-score of 97.84% with a false-positive rate of 0.138%, outperforming a 1-D CNN baseline, which achieves an F1-score of 92.01% and a false-positive rate of 3.388%, and a Random Forest baseline, which achieves an F1-score of 67.29%. Welch’s t-tests confirm statistically significant differences in monthly F1-scores between the proposed approach and both baselines. Moreover, MalBERT-Temporal maintains monthly F1-scores within a 2.2-point band throughout the nine-month evaluation period, indicating improved robustness to temporal distribution shift. A component-wise ablation traces the improvement to the tokenized self-attention mechanism itself: substituting a token-wise feed-forward block for self-attention while keeping every other component costs 2.73 F1 points, and removing the tokenization costs 1.85 points, while positional embeddings, the CLS token, multi-scale pooling, encoder depth, and nonlinearity are each worth 0.47 points or less. A matched random-split control that leaves the model, the preprocessing, the training budget, and the calibration procedure unchanged gives an F1-score of 98.86%, which confirms that chronological evaluation is the stricter protocol.

Zero-day malware detection transformer self-attention portable executable temporal isolation concept drift BODMAS structural feature modeling

Abstract doi.org/10.14569/IJACSA.2026.0170820 PDF

21

Next-Generation Explainable Sensor-Fusion-Based Artificial Intelligence Framework for In-Home Diabetic Foot Ulcer Risk Prediction

Author 1: Jayashree J Author 2: Vijayashree J Author 3: Vijayarajan V Author 4: Saravanan S

Diabetic Foot Ulcers (DFUs) are one of the most serious but preventable complications of diabetes mellitus that often develop without clinical signs until the late stages. This means that neuropathic and vascular abnormalities should be diagnosed in time to be effectively prevented. This study presents the design and development of a home-based smart monitoring system that can be used to identify DFU risk in its early stage by non-invasive sensor-fusion-based monitoring. A baseline version of the system with a plantar pressure sensor and a vibration-based tactile sensitivity evaluation was developed to detect early neuropathic changes in foot health with the use of a composite learning model. Building on this baseline, the proposed framework extends the sensor array with force-sensing resistors (FSRs) to map abnormal plantar pressure regions, vibration motors to evaluate tactile sensitivity, DS18B20 temperature sensors to evaluate dermal thermal variation, and a pulse sensor to evaluate the plantar blood flow. SHAP (SHapley Additive exPlanations) guided RFE (Recursive Feature Elimination) is used to process sensor signals to identify physiologically salient features. One of these is a hybrid Multi-Layer Perceptron-Bidirectional Long Short-Term Memory (MLP-BiLSTM) model with a fuzzy inference layer that then categorizes DFU risk into Low, Medium, and High. Explainable Artificial Intelligence (XAI) methods, such as SHAP and LIME (Local Interpretable Model-agnostic Explanations), are incorporated to increase clinical understandability. Experimental results indicate that the proposed model improves both predictive performance and explainability relative to the baseline methods evaluated, supporting consistent, clinician-interpretable home-based screening of diabetic foot risk within the studied population; broader external validation across additional sites remains necessary before wider clinical generalization can be claimed.

DFU risk assessment sensor-fusion-based monitoring system Explainable Artificial Intelligence (XAI) MLP-BiLSTM hybrid model home-based healthcare monitoring

Abstract doi.org/10.14569/IJACSA.2026.0170821 PDF

22

ReadTech: A Mobile Reading Application for Grade 7 Learners

Author 1: Mary Jane U. Quibilan

Reading comprehension remains a fundamental skill for academic success; however, many Grade 7 learners continue to experience difficulties in vocabulary development, text comprehension, and access to engaging instructional materials. This study developed and evaluated the ReadTech Mobile Reading Application using a Design and Development Research (DDR) methodology. The study integrated developmental, descriptive-survey, and evaluative approaches involving Grade 7 learners, expert validators, and student-users from two integrated schools in Asingan, Pangasinan, Philippines. Needs assessment results revealed that learners had a strong need for supplementary reading materials, vocabulary-building activities, and technology-supported learning resources. The developed application incorporates personalized learning activities, reading lessons, assessments, and progress monitoring features aligned with the Grade 7 English MATATAG Curriculum. Expert evaluation indicated that the application was highly valid (M = 4.71), particularly in task accuracy and technology usability. Likewise, student-users rated the application as highly acceptable (M = 4.47), with engagement receiving the highest rating. These findings indicate that ReadTech is a curriculum-aligned mobile learning application that was successfully developed, validated by experts, and positively received by learners. The study demonstrates the feasibility of using Design and Development Research (DDR) to develop learner-centered mobile applications that support reading instruction and provide a foundation for future studies evaluating learning outcomes.

Android application digital learning educational technology mobile learning reading comprehension

Abstract doi.org/10.14569/IJACSA.2026.0170822 PDF

23

A Blockchain-Based Role-Based Access Control Gateway for Secure Electronic Health Record Access and Tamper-Evident Auditing

Author 1: Osamah Saleh Author 2: Shouki A. Ebad

Electronic Health Record (EHR) systems store sensitive patient information and require strong access control and reliable audit records. Traditional centralized Role-Based Access Control (RBAC) systems may be vulnerable to unauthorized access, privilege escalation, audit-log modification, and undetected changes to stored records. This study developed a lightweight blockchain-based RBAC gateway integrated with OpenMRS. The system used three roles, Admin, Doctor, and Patient, and one Solidity smart contract to manage role assignments, Doctor–Patient permissions, access decisions, trusted record hashes, and blockchain audit events. A Node.js gateway acted as the policy-enforcement point, while SQLite was used as a centralized audit-log baseline. The system was evaluated through authorized-access, unauthorized-access, privilege-escalation, audit-log tampering, EHR data-tampering, gas-use, and throughput experiments. All 50 measured authorized requests were allowed and completed the expected OpenMRS retrieval workflow, while all 50 unauthorized-access attempts were blocked before reaching OpenMRS. The SQLite decision could be changed from deny to allow, while the corresponding blockchain event remained unchanged. The trusted-hash experiment detected all 10 simulated EHR modifications. Median end-to-end latency was 31.8 ms for denied requests and 468.9 ms for authorized requests. Sequential smart contract throughput averaged 15.0850 TPS for authorized access and 15.4214 TPS for unauthorized access. These findings show that the proposed gateway can improve access-control enforcement, detect off-chain record changes, and provide tamper-evident audit evidence without storing full EHR records on-chain.

Electronic health records role-based access control blockchain smart contracts OpenMRS audit logs access control cybersecurity

Abstract doi.org/10.14569/IJACSA.2026.0170823 PDF

24

A Cross-Sector Leadership Framework for Organizational Change and Sustainability: Testing the Impact of Intrinsic Leadership Values Using Structural Equation Modeling

Author 1: Sudirman Said Author 2: Aurik Gustomo Author 3: Yudo Anggoro

Purpose: Leading involves challenging the status quo. A critical function of leadership is to drive change, ensuring both the immediate performance and long-term sustainability of organizations. Design/methodology/approach: Research on leadership, changes, and sustainability has focused predominantly on individual sectors—public, private, or non-profit. Despite the increasing demand for cross-sectoral collaborations, studies that encompass all three sectors—public, private, and non-profit organizations—are relatively scarce. Previous qualitative research has identified five Intrinsic Leadership Values (ILV), each supported by specific characteristics, and developed a unified leadership framework applicable across organizations in public, private, and non-profit sectors. Findings/results: These studies also revealed a universal pattern for managing changes that applies to organizations of all types. To understand the connections between ILV, Organizational Change (OC), Sustainable Change (SC), and Organizational Sustainability (OS), the research developed a framework that was analyzed through case studies. Practical implications: Building on the framework identified through earlier qualitative studies, this research uses a quantitative methodology to examine its generalizability. A purposive sampling survey of 580 respondents, holding managerial positions in public, private, and non-profit organizations in Indonesia, was conducted. Originality/value: Using Structural Equation Modeling (SEM) and Analysis of Moment Structures (AMOS) as analytical tools, this research corroborates previous findings: ILV has a significant impact on OC, which, in turn, enhances OS, while ILV has no influence on SC. It also confirms the reciprocal relationship among these constructs, with ILV serving as the pivotal unifying variable.

Intrinsic leadership values organizational change sustainable change organizational sustainability leadership development cross-sector leadership public-private-nonprofit collaboration change management

Abstract doi.org/10.14569/IJACSA.2026.0170824 PDF

25

Research Trends in Human-Computer Interaction for Immersive Virtual Reality Applications: A Bibliometric Study (2010-2026)

Author 1: Nur Sauri Yahaya Author 2: Sobihatun Nur Abdul Salam Author 3: Hanis Salwani Othman

The rapid advancement of immersive technologies has significantly expanded the scope of Human-Computer Interaction (HCI) within the context of Immersive Virtual Reality (IVR). Despite the growing body of research, the field remains fragmented, with limited understanding of its intellectual and thematic development. This study conducts a comprehensive bibliometric analysis of HCI research within IVR using data retrieved from the Web of Science (WoS) Core Collection. A total of 1,185 publications from 2010 to 2026 were initially identified and refined to 1,129 records following a systematic screening process. The analysis employs co-citation and keyword co-occurrence techniques using VOSviewer to examine both the knowledge base and thematic development of the field. The findings reveal a substantial increase in publication output and citation impact, indicating the growing maturity and relevance of HCI research in IVR. Co-citation analysis identifies four major intellectual domains: presence and immersive experience, user experience and application contexts, usability and human factors, and behavioral and psychological interaction. Meanwhile, keyword co-occurrence analysis uncovers four dominant thematic clusters, including core interaction and system design, human-centered computing, visualization and spatial interaction, and emerging immersive technologies. This study contributes by providing a structured and data-driven mapping of the research landscape, emphasizing the relationships between interaction design, user experience, and technological advancement. The findings offer both theoretical and practical implications for advancing more integrated and user-centered approaches in immersive virtual environments, while also identifying key directions for future research.

Human-computer interaction (HCI) Immersive virtual reality (IVR) immersive technologies User experience (UX) bibliometric analysis science mapping

Abstract doi.org/10.14569/IJACSA.2026.0170825 PDF

26

Memory-Augmented Autoencoder for Industrial Anomaly Detection

Author 1: Hai Son Author 2: Duc Ngo

Unsupervised anomaly detection in industrial manufacturing requires identifying defective products utilizing only defect-free training images. Reconstruction-based methods using autoencoders are widely applied but suffer from limited discriminative power when defects occupy small image regions. This work introduces a Memory-Augmented Autoencoder (MAA) that combines a U-Net reconstruction component with a patch-level memory bank for nearest-neighbor-based anomaly scoring. During inference, MAA computes a hybrid anomaly score from both the reconstruction error and the memory distance to produce image-level predictions and spatial anomaly heatmaps. Evaluated on the bottle category of the MVTec AD benchmark, MAA attains an image-level AUC-ROC of 0.985 and an Average Precision of 0.996, surpassing all autoencoder-based baselines including SSIM AE (0.930) and VAE (0.910) and performing on par with the pretrained PaDiM method using a ResNet-18 backbone (0.982), while relying on only 168 defect-free training images and no ImageNet pretraining. A pixel-level AUC-ROC of 0.895 indicates a reasonable spatial localization capability. Ablation studies demonstrate that the memory-based component is the primary driver of detection performance, outperforming the reconstruction component by 16.4 percentage points in AUC-ROC, which is attributable to the robustness of patch-level memory matching for localized structural defects. As the present evaluation is confined to a single object category, the reported results are best interpreted as a proof of concept rather than as evidence of broad generalization.

Anomaly detection autoencoder memory bank nearest-neighbor search MVTec AD unsupervised learning industrial inspection U-Net

Abstract doi.org/10.14569/IJACSA.2026.0170826 PDF

27

Understanding Player Choice in Digital Games: Qualitative Study Using Thematic Analysis

Author 1: Faris Haziq Sarif Author 2: Amelia Jati Robert Jupit Author 3: Jacey-Lynn Minoi Author 4: A. Imran Nordin

Player choice is a fundamental aspect of digital games, shaping how players interact with virtual worlds and narratives. In many games, player choices are reflected through mechanics such as morality systems, which record decisions as virtuous or harmful and shape how characters and the game world react. For game designers, player choice is vital, granting players freedom to shape their experiences and fostering deeper engagement. When players are presented with significant choices, they are more likely to remain invested and prolong their gameplay. Despite its importance, research on player choice remains limited, with existing research often focusing on challenge, immersion, or motivation within specific genres while overlooking broader implications. We conducted interviews to explore the factors that influence player choice in digital games. The data were examined using thematic analysis to identify recurring patterns and key themes. Results from the interviews revealed six main dimensions of player choice: Goals, Character Personalization, Gameplay, Social Interaction, Rewards, and Completion, each with relevant sub-themes.

Player choice thematic analysis player engagement

Abstract doi.org/10.14569/IJACSA.2026.0170827 PDF

28

Metaheuristic-Based Test Case Selection for Regression Testing: A Systematic Literature Review

Author 1: Siti Hawa Mohamed Shareef Author 2: Rabatul Aduni Sulaiman Author 3: Nazri Mohd Nawi Author 4: Wan Noor Hamiza Wan Ali Author 5: Nurul N. Jamal Author 6: Fairuz Amalina

Regression testing plays a crucial role in ensuring that software modifications do not adversely affect existing functionality. However, test suites continue to grow, and the retest-all method has become increasingly impractical due to high execution costs and time constraints. Consequently, Test Case Selection (TCS) has emerged as an important optimization method to identify which test cases are relevant for re-execution. Metaheuristic optimization has received a great deal of attention in Search-Based Software Testing (SBST) for balancing the conflicting objectives of cost reduction and fault detection effectiveness. However, there are still multiple areas of fragmentation within the research, including algorithm design, objective formulation, empirical evaluation, and reporting practices. Results indicate that evolutionary algorithms and swarm intelligence are the primary algorithms used for TCS, with an increasing trend towards hybrid and multi-objective approaches to increase the quality and scalability of the search. Multi-objective formulations are significant in this context, as such approaches provide a structured mechanism to capture the trade-off between testing cost and effectiveness through Pareto-based evaluations. Empirical evidence remains largely dependent on benchmark and open-source systems, although recent studies increasingly incorporate industrial and Artificial Intelligence (AI)-based environments. Overall, the findings indicate that there is a lack of single techniques that are universally superior, as performance is strongly influenced by system characteristics and evaluation criteria. Future research should emphasize standardized evaluation practices, stronger industry-scale validation, and the development of domain-aware TCS techniques to improve practical applicability.

Regression testing test case selection metaheuristic algorithms search-based software testing

Abstract doi.org/10.14569/IJACSA.2026.0170828 PDF

29

TriMFB: A Tri-Feature Fusion Framework Integrating Modality-Specific and Cross-Modal Representations for Fine-Grained Fake News Detection

Author 1: Idza Aisara Norabid Author 2: Masita Jalil Author 3: Noor Hafhizah Abd Rahim

The rapid spread of multimodal misinformation on social media platforms has strengthened the need for effective fake news detection systems. While recent multimodal approaches integrate textual and visual information, most existing studies formulate the task as binary classification, thereby failing to fully exploit the rich label structures available in fine-grained datasets. Moreover, conventional fusion strategies such as simple concatenation often fail to capture complex cross-modal interactions. To address these limitations, this study proposes TriMFB, an attention-enhanced Multimodal Factorized Bilinear (MFB)-based early fusion framework for fine-grained multimodal fake news detection. Unlike other frameworks, the proposed TriMFB simultaneously integrates modality-specific features, attention-refined features, and MFB-based cross-modal interactions within a unified tri-feature early fusion architecture. The proposed model extracts textual and visual representations using Bidirectional Encoder Representations from Transformers (BERT) and a 50-layer Residual Network (ResNet50), respectively. Modality-specific features are refined through attention mechanisms and integrated using MFB pooling to capture cross-modal interactions. Experiments conducted on the Fakeddit dataset demonstrate that the proposed framework outperforms baseline bilinear fusion models, achieving a classification accuracy of 0.78 in a six-class setting. Although the overall accuracy remains comparable, the proposed tri-feature fusion strategy substantially improves the detection of challenging minority classes, increasing the F1-score for the Manipulated class from 0.21 to 0.29. These findings suggest that this fusion strategy offers a practical route to more reliable detection of subtly manipulated content, a minority class that is both harder to detect and disproportionately harmful when missed by content moderation systems.

Bilinear pooling early fusion fake news detection fine-grained multimodal

Abstract doi.org/10.14569/IJACSA.2026.0170829 PDF

30

An Intelligent On-Demand Laundry Logistics Platform with Real-Time Location Tracking and Machine Learning-Based Customer Support

Author 1: Muhammad Faris Rosli Author 2: Siti Zuraidah Ibrahim Author 3: Mohd Nazri A Karim Author 4: Mohd Hafizuddin Mat Author 5: Faridah Hanim Mohd Noh Author 6: Tanakorn Inthasuth

Traditional on-demand laundry operations suffer from inefficient manual coordination, a lack of real-time delivery visibility, and delayed customer support handling. To address these operational challenges, this study presents LaundrOTrack, an intelligent on-demand laundry logistics platform connecting customers, drivers, and administrators within a unified digital ecosystem. Powered by Flutter for cross-platform access and Firebase Cloud Firestore for real-time data synchronization, the system incorporates Location-Based Services (LBS) via the Google Maps API to streamline last-mile delivery workflows. The platform's real-time location tracking system utilizes continuous GPS updates, route polyline visualization, dynamic Estimated Time of Arrival (ETA) calculation, and spatial arrival boundary detection to monitor pickup and return deliveries. Furthermore, machine learning-based customer support is integrated through an automated conversational agent (Laundra Ask). The underlying natural language processing pipeline employs text preprocessing, Term Frequency-Inverse Document Frequency (TF-IDF) feature extraction, and a Linear Support Vector Classifier (Linear SVC) model hosted on a Flask REST API to predict user query intents and deliver context-aware responses. System testing demonstrates seamless multi-role data synchronization, precise geolocation triggering, and high-accuracy query intent classification. The proposed platform significantly improves order transparency, reduces manual administrative overhead, and optimizes end-to-end service efficiency in smart service logistics.

On-demand laundry logistics Location-Based Services (LBS) linear SVC intent classification flutter cloud firestore

Abstract doi.org/10.14569/IJACSA.2026.0170830 PDF

31

Learning to Schedule Machines and Operators: A Human-Aware Deep Q-Learning Framework for Dynamic Flexible Job Shops

Author 1: Taji Hajar Author 2: Ayad Ghassane Author 3: Zaki Abdelhamid Author 4: Khammal Adil

Most learning-based schedulers for job shop problems assume static workforce performance. However, dynamic dual-resource job shops must navigate stochastic disturbances and time-varying task durations due to human factors. This study proposes a human-centered Deep Reinforcement Learning framework, DD4LQN, for dynamic flexible job shop scheduling under operator learning and forgetting dynamics. The environment integrates a disturbance-aware scenario generator, bounded logistic learning-forgetting dynamics, and deterministic Dual Pair Ranking to ensure auditable decisions. The training used a plan that included ENTRY and EXIT greedy evaluations and a Quote-then-Commit process. Tests over ten weeks showed that DD4LQN-EXIT did better than other scheduling methods. Empirical evaluations over a ten-week horizon demonstrate that DD4LQN-EXIT got the average schedule reward of 0.5843. It was better than Earliest Due Date (EDD) and Shortest Processing Time (SPT) by 17% and 14%, respectively. It also completed jobs and had fewer risky orders. Even though Shortest Processing Time had delays on average, it completed fewer jobs because it focused on short tasks. Additionally, an exploratory ablation across the same scenarios demonstrated that Dual Pair Rank and learning–forgetting dynamics come up with better results. Furthermore, representation diagnostics on the 227,717-parameter network confirm structural stability, with layer matrices retaining up to 98.2% of maximum effective rank without capacity collapse.

Deep reinforcement learning dynamic flexible job shop dual-resource scheduling human-centered manufacturing learning and forgetting

Abstract doi.org/10.14569/IJACSA.2026.0170831 PDF

32

Institutional Readiness for Generative AI-Enhanced Assessment and Feedback in Higher Education: Evidence from an Exploratory Study

Author 1: Dolantina Hyka Author 2: Elion Shabanaj Author 3: Jurgen Mecaj Author 4: Ardita Hykaj Author 5: Elton Skendaj

The emergence of Generative Artificial Intelligence (GenAI) technology is changing the ways of assessment and feedback procedures in higher education institutions by allowing for a more flexible and personalized process of learning. However, the effective use of GenAI requires pedagogic design, governance, and implementation strategies. The current study will discuss the institutional prerequisites to implement GenAI in assessment and feedback procedures (Activity A2.2 of the Erasmus+ HEGenAI project). Empirical data were gathered using structured questionnaires completed by educators (n=61) and students (n=254). The descriptive approach involving frequencies, percentages, means, and standard deviations helped to investigate current usage of AI-supported assessment procedures, institutional requirements for GenAI adoption, ways of implementing GenAI, educational advantages of GenAI, and associated risks. The results showed that while GenAI is used extensively to facilitate assessment-related processes, the usage remains predominantly informal and non-institutionalised. It is argued that there is a need to use licensed GenAI solutions, to train educators, to redesign assessment processes, and also to develop an appropriate governance framework. Academic integrity, critical thinking, and AI dependency emerged as the most important risks requiring human attention.

Generative AI assessment feedback higher education academic integrity formative assessment institutional readiness

Abstract doi.org/10.14569/IJACSA.2026.0170832 PDF

33

Earthquake Damage Risk Classification Using Machine Learning Models Based on Built-Up Area Indices from Satellite Imagery

Author 1: Gunawan Prayitno Author 2: Eko Sediyono Author 3: Irwan Sembiring Author 4: Sri Yulianto Joko Prasetyo

One of the challenges in assessing earthquake damage risk in areas with high seismic activity and limited data is the lack of a detailed building inventory and the absence of available data. Therefore, a remote sensing and machine learning framework is needed that can utilize the built-up area index with NDBI from multitemporal Landsat 8 imagery as a vulnerability proxy, combined with ISGS seismic hazard data, topographic variables from DEM, and surface geology. Support Vector Machine (SVM) and Random Forest (RF) classifiers can be trained and developed to categorize vulnerability classes and evaluate results in terms of accuracy, precision, recall, and F1 score. The SVM achieved an accuracy of 0.970, with precision, recall, and F1 scores all reaching 0.97, while the RF achieved 0.96. This study focuses on Nabire Regency, Central Papua, Indonesia, utilizing various approaches, scaling methods, and a relatively low-cost methodology for the initial screening of earthquake-prone areas without the need for field data collection, with the potential to apply these methods to other seismic regions.

Seismic risk assessment vulnerability proxy remote sensing machine learning data scarcity

Abstract doi.org/10.14569/IJACSA.2026.0170833 PDF

34

Semantic Interoperability of Heterogeneous Information Systems: A Hybrid Framework for Machine Learning-Assisted Semantic Mapping Recommendation

Author 1: Aïcha KOULOU Author 2: Mohamed CHEKOUR Author 3: Norelislam EL HAMI Author 4: Nabil HMINA

Semantic interoperability is a major challenge in the integration of heterogeneous information systems, where differences in data structures, terminologies, and representations complicate the automatic identification of schema matches. Traditional schema matching approaches, primarily based on rules or lexical similarity measures, have limitations in the face of the increasing complexity of digital environments. This article proposes a hybrid framework to improve the automatic recommendation of semantic mappings by combining lexical, structural, statistical, semantic, and business similarities with a supervised learning model. The framework also incorporates expert validation to ensure the quality of the recommended matches and to feed an iterative process for improving the mapping repository. The evaluation is based on a case study in the health insurance sector. Experimental results show that the XGBoost model achieves the best performance with an F1 score of 89.48% and an AUC of 0.915. Top-k evaluations also highlight excellent recommendation capabilities, with a Hit@1 of 83.51%. The ablation study confirms that combining different similarity families significantly improves recommendation quality. These results demonstrate the value of the proposed framework for enhancing semantic interoperability, facilitating the identification of matches between heterogeneous schemas, and reducing the validation effort required from domain experts.

Semantic interoperability schema matching machine learning XGBoost information systems

Abstract doi.org/10.14569/IJACSA.2026.0170834 PDF

35

Can We Trust Concept-Based ECG Explanations? A TCAV Analysis of Confounding and Reproducibility in Atrial Fibrillation Detection

Author 1: El Hassane Jennah Author 2: Houda Indjaren Author 3: Lhoucine Ben Taleb Author 4: Hamid El Malali Author 5: Azeddine Mouhsen

AI explanation methods are typically validated in one of two ways: by confirming that an explanation is not an artifact of the underlying data (a confound check), or by confirming that it generalizes to unseen data. Fewer studies examine whether an explanation is stable across independently trained copies of the same model. We present a validation framework combining both dimensions — confound control and model-seed reproducibility — and demonstrate it in a case study using a deep learning model trained to detect atrial fibrillation (AF) from long, continuous ECG recordings. Using concept activation vectors (CAVs), we test whether the model's predictions are associated with three clinically meaningful patterns: irregular heartbeat timing, the presence of abnormal beats, and proximity to a rhythm change. We find that a concept-prediction association can be statistically significant and survive a targeted AF-label confound control in one trained model, yet reverse direction when the same architecture is retrained from a different random initialization. Across the nine concept-depth combinations examined, three replicated consistently across five independently trained models, two were inconclusive, and four showed evidence of directional instability under the predefined criterion. These findings demonstrate, within this AFNet/LTAFDB case study, that statistical significance and confound robustness alone may not establish a stable concept-based explanation; model-seed reproducibility provides a complementary validation dimension that can reveal sensitivity to training initialization.

Explainable AI concept activation vectors TCAV atrial fibrillation reproducibility electrocardiogram

Abstract doi.org/10.14569/IJACSA.2026.0170835 PDF

36

HR-CD-Auth: Consent-Bound Cross-Domain Pseudonymous Authentication for Vocational Counseling and Employment Matching

Author 1: Haewon Byeon Author 2: Changmin Keum

Vocational counseling platforms move sensitive records across counseling centers, enterprise human-resource systems, public employment services, and automated matching engines. Direct reuse of cross-domain authentication leaves identifiers linkable, separates authentication from purpose and consent, and gives revocation no protocol-level effect on cached credentials. HR-CD-Auth combines domain-scoped pseudonyms, compartmented data keys, consent-bound capabilities, and an ephemeral X25519 handover. Automated analysis of the initial capability-only fast path with ProVerif 2.05 found four failures: protected-record secrecy, accepted-session-key secrecy, and both injective-authentication correspondences. The repaired protocol delivers a fresh one-time handover key in separately protected client and gateway tickets, authenticates the first handover message, and mixes that key with the X25519 secret in HKDF. All four ProVerif queries then returned true. An OpenSSL-backed Python proof of concept executed three independent runs of 10,000 in-memory handovers on an AMD Ryzen 7 8845HS. Run-level medians were 0.2773-0.2812 ms, p95 latency was 0.4669-0.4820 ms, sequential throughput was 3041.5-3119.9 handovers/s, and the three online messages totaled 444 bytes. These figures isolate the cryptographic path and exclude network, TLS, database, policy-engine, and ledger delay. Identity resolution is separated from the online trust service through a two-authority, 2-of-2 threshold profile. The result is a measured and symbolically checked design whose remaining deployment and governance assumptions are explicit.

Anonymous authentication consent management cross-domain authentication employment matching forward secrecy pseudonymization vocational counseling

Abstract doi.org/10.14569/IJACSA.2026.0170836 PDF

37

Popularity Bias and Category Diversity in Review-Based Rating Prediction for E-Commerce: An Empirical Study on Indonesian Marketplace Data

Author 1: Andy Supriyadi Author 2: Herman Dwi Surjono Author 3: Handaru Jati

Online marketplaces increasingly rely on user reviews to estimate product quality. However, most empirical comparisons of review-based prediction models report only aggregate accuracy and overlook how item popularity and category composition shape that accuracy. This study presents a controlled empirical analysis of review-based rating prediction on a self-collected, anonymized corpus of 44,229 Indonesian marketplace reviews spanning four product categories (fashion, electronics, tools/hardware, and sports). To isolate the effect of category diversity from data volume, the experimental design was set with the total number of reviews kept constant at 12,000, while the number of categories was increased from two to four. For product metadata classification, three text classification models (TF-IDF with logistic regression, linear SVM, and random forest) and a customized IndoBERT transformer model were compared using five-fold cross-validation, and the difficulty of prediction was further analyzed at the item level. Three findings emerge. First, the review text is the dominant signal, reducing the mean absolute error to 0.62 (0.57 with IndoBERT), compared with 1.28 for the majority baseline. Second, increasing category diversity within a fixed budget results in a small but consistent performance degradation. Third, and most notably, popular items are systematically harder to predict than long-tail items within every category, and per-item error correlates positively with rating variance and sales volume and negatively with average product rating and price. The polarization of evaluations is considered a major contributing factor to production difficulties; the protocol can be reproduced and interpreted at a level where evaluation predictions are based on a pass/fail assessment.

Review-based rating prediction popularity bias category diversity product characteristics e-commerce Indonesian text classification

Abstract doi.org/10.14569/IJACSA.2026.0170837 PDF

38

Orientation-Aware Feature Fusion for Accurate Rotated Object Detection in Remote Sensing Images

Author 1: Jing Zhang Author 2: Mas Rina Mustaffa Author 3: Fatimah Khalid Author 4: Zainal Abdul Kahar

Conventional feature fusion mechanisms largely overlook orientation information, making it difficult to effectively represent objects with diverse rotational patterns. To address this issue, we propose YOLO-RSL, a lightweight rotated object detector that introduces orientation awareness into feature representation, feature fusion, and localization optimization. The proposed Rotation-Oriented Iterative Attentional Feature Fusion (RO-iAFF) module incorporates implicit orientation information derived from orthogonal asymmetric depthwise convolutions into a two-stage channel attention process, enabling orientation-sensitive multi-scale feature aggregation. In addition, a Swin Transformer block and a large-kernel feature branch are employed to enhance global context modeling and improve small-object representation. To achieve more accurate localization, an enhanced regression loss (LFFE Loss) is designed to jointly optimize overlap quality, rotation angle, center offset, and aspect ratio. Experimental results on the DIOR-R and UCAS-AOD datasets show that YOLO-RSL achieves mAP0.5 scores of 84.17% and 98.44%, surpassing the baseline by 3.15% and 2.91%, respectively. Moreover, the performance gains become more pronounced as target aspect ratios increase, demonstrating the effectiveness of the proposed orientation-aware design.

Rotated object detection oriented bounding box feature fusion remote sensing image

Abstract doi.org/10.14569/IJACSA.2026.0170838 PDF

39

Feasibility of Generative Artificial Intelligence and Large Language Model Adoption to Enhance Organisational Knowledge Management Practices

Author 1: Surya Sumarni Hussein Author 2: Nur Azaliah Abu Bakar Author 3: Saidatul Rahah Hamidi Author 4: Shuhaida Mohamed Shuhidan Author 5: Siti Salbiyah Abdul Geni Author 6: Anitawati Mohd Lokman

Generative artificial intelligence (Gen AI) and large language models (LLMs) offer substantial potential to improve how organisations capture, organise, retrieve and reuse knowledge. Existing knowledge management (KM) frameworks, however, seldom integrate Gen AI/LLM-specific processes, data governance, and ethical requirements in highly regulated public-sector settings, and few have been empirically evaluated. This study assesses the feasibility of adopting GenAI and LLMs for organisational KM and develops a corresponding Knowledge Management and Artificial Intelligence (KMAI) framework for a national communications regulator. Guided by Diffusion of Innovation (DOI) theory, the study employed a multi-method qualitative design comprising a literature review, semi-structured interviews with six KM representatives, a focus group discussion involving fourteen participants using the LEIQ™ model for SWOT analysis, and a content-validity evaluation by three experts. Interview data were analysed thematically, SWOT findings were transformed into strategies through TOWS analysis, and framework relevance was assessed using item- and scale-level content validity indices. The findings indicate that adoption is feasible in this setting: participants perceived clear relative advantage, compatibility, trialability and observability, while complexity was manageable when supported by adequate skills, data classification, secure infrastructure and governance. The principal risks concerned data quality, privacy, security, misinformation, over-reliance on AI, and organisational resistance. The proposed KMAI framework achieved acceptable content validity across all clusters, with S-CVI/Ave values ranging from 0.89 to 1.00, and was refined into four strategic thrusts: People, Process, Technology and Data. Because the evidence derives from one organisation and a three-member expert panel, the framework is presented as an empirically grounded and internally validated proposition whose extension to other agencies is analytic rather than statistical; the boundary conditions governing such extension are stated explicitly, and confirmatory validation with a larger expert panel and independent organisational sites is identified as the necessary next step. The study extends DOI-based adoption analysis by showing that data governance and ethics condition Gen AI/LLM adoption in organisational KM and provides a practical, staged framework for regulated public-sector organisations.

Generative artificial intelligence (Gen AI) large language models (LLM) knowledge management diffusion of innovation (DOI) lokman's emotion and importance quadrant (LEIQ™) content validity

Abstract doi.org/10.14569/IJACSA.2026.0170839 PDF

40

Machine Learning for Distributed Acoustic and Fibre-Optic Sensing in Infrastructure Monitoring: A Systematic Review (2015-2025)

Author 1: Abd Rahim Saidin Author 2: Illani Mohd Nawi

Distributed acoustic sensing (DAS), together with broader distributed fibre-optic sensing (DFOS), has emerged over the past decade as a candidate technology for large-scale infrastructure monitoring, while machine learning (ML) is increasingly used to interpret its high-throughput data. This study presents a PRISMA 2020-guided bibliometric and systematic-mapping review of this convergence. A structured search of Web of Science, Scopus, and IEEE Xplore returned 291 records published between 2015 and 2025; after duplicate removal and two stages of screening, a final analytical corpus of 31 publications was established, of which 16 are primary DAS-with-ML infrastructure studies. Publication output rises sharply to seven papers in 2024 and twelve in 2025. Among the primary studies, convolutional neural networks (CNNs), recurrent models, and CNN-LSTM hybrids are prominent, with more recent use of few-shot learning, graph neural networks (GNNs), generative models, and self-supervised learning. Perimeter security is the most represented application domain, followed by traffic, railway, and structural-health monitoring. Critical synthesis shows that the apparent dominance of CNNs is partly explained by the natural two-dimensional spatiotemporal representation of DAS data and by mature, reusable CNN pipelines, rather than evidence of universal superiority. Cross-study comparison remains limited by heterogeneous sensing environments, inconsistent acquisition and label reporting, non-uniform evaluation protocols, sparse computational reporting, and minimal long-term robustness evaluation. Five research gaps are identified: cross-infrastructure generalisation, standardised benchmarks and open datasets, physics-data-driven integration, interpretability and uncertainty, and operationally robust edge deployment.

Bibliometric analysis convolutional neural network deep learning distributed acoustic sensing distributed fibre-optic sensing infrastructure monitoring machine learning phase-sensitive OTDR structural health monitoring systematic review

Abstract doi.org/10.14569/IJACSA.2026.0170840 PDF

41

CEM MoE: A Lightweight Sparse Mixture of Experts Framework for Empathetic Dialogue Generation

Author 1: Zhinan Gou Author 2: He Liu Author 3: Yufan Wang

Empathetic dialogue generation requires models to understand user emotions and produce fluent, relevant, and emotionally appropriate responses. Existing lightweight models offer favorable deployment efficiency, but shallow emotion prediction layers limit their ability to represent fine-grained and dynamically changing emotional states. To address this limitation, CEM-MoE, a lightweight sparse mixture of experts (MoE) framework based on the Commonsense-aware Empathetic Chatting Machine (CEM), is proposed. Unlike the original CEM, CEM-MoE employs two independently routed sparse MoE streams to separately model emotional information from dialogue history and the current utterance. A bounded density-aware fusion mechanism integrates the two emotion distributions, where normalized utterance length serves as a lightweight density proxy and a sigmoid gate constrains the fusion coefficient to a valid range. Auxiliary load balancing regularization and label smoothing further improve expert utilization and generation stability. Experiments on the EmpatheticDialogues and ESConv datasets show that CEM-MoE achieves competitive generation performance with only marginal additional overhead. Compared with CEM, CEM-MoE reduces PPL from 37.76 to 36.59 and improves emotion accuracy from 32.10% to 35.47%, while increasing the parameter count by only 0.12M. Expert activation analysis reveals no obvious expert collapse, and human evaluation indicates higher scores in empathy, fluency, and relevance. The results verify that the dual-stream sparse MoE with density-aware fusion is effective for lightweight fine-grained empathetic dialogue modeling. The source code is publicly available at https://github.com/ZhinanGou/CEM-MoE.

Empathetic dialogue generation sparse mixture of experts lightweight dialogue model emotion modeling expert routing

Abstract doi.org/10.14569/IJACSA.2026.0170841 PDF

42

From Personality to Performance: An Automated Grouping Tool for Collaborative Learning

Author 1: Dalila Raissouni Author 2: Ahmed El Oualkadi Author 3: Kamal Reklaoui

As part of pre-service teacher training, this study presents a new web-based tool that intelligently groups learners based on their personality profiles (PCM, or Process Communication Model) to enhance collaborative work in flipped classrooms. The software automatically groups users by personality based on the results of a personality test. The application is compatible with all mobile devices and can be accessed through a web browser without the need for installation. The algorithm aims to improve role distribution, interpersonal communication, and conflict resolution by maximizing diversity within each group while taking contextual factors into account. Inspired by popular educational platforms like Kahoot, the user interface provides an intuitive experience for both teachers and students. Teachers can create classes, set group sizes, monitor results, and create ideal groupings with little work. A quasi-experimental pre–post implementation involving 267 trainee teachers, with a response rate of 63% (164 participants), revealed statistically significant improvements in collaboration and participant involvement following group reorganization through the application. These findings support the pedagogical relevance of integrating personality-based grouping technologies. They suggest that such technologies may foster learner engagement and self-awareness, and that they are associated with improved collaboration in educational environments. They suggest that such technologies may foster learner engagement, self-awareness, and effective collaboration in educational environments. The study contributes to the field by providing empirical evidence of the applicability of the PCM framework in teacher training contexts and by demonstrating the feasibility of integrating personality-based grouping mechanisms into digital learning environments.

Personality-based grouping Process Communication Model (PCM) Computer-Supported Collaborative Learning (CSCL) team diversity educational technology teacher training flipped classroom collaborative learning

Abstract doi.org/10.14569/IJACSA.2026.0170842 PDF

43

Development of An Artificial Neural Network-Based System to Detect Lane and Roadside Traffic Signs

Author 1: Viraj Sonawane Author 2: Balasaheb Agarkar Author 3: Sachin Chaudhari

Ensuring road safety requires timely and accurate detection of lanes and roadside traffic signs, which are essential for assisting drivers and reducing accident risks. However, robust simultaneous detection remains challenging due to variations in illumination, occlusions, complex backgrounds, and diverse road conditions. This research introduces a novel vision-based system for lane detection and roadside traffic sign recognition using advanced artificial neural network architectures. In the pre-processing stage, Contrast Limited Adaptive Histogram Equalization (CLAHE) is applied to enhance image visibility under varying lighting conditions. For classification, a Self-Artificial Attentive Gated Neuro-Recurrent Unit (SAAG-NRU) is proposed, a novel architecture integrating deep artificial neural network (ANN) layers, Gated Recurrent Units (GRU), and a self-attention mechanism, enabling sequential refinement of features and prioritization of the most safety-relevant cues. In addition, YOLOv11 was utilized to train and detect traffic signs exclusively. For lane detection, a lane mask-based approach is implemented to accurately segment and highlight lane boundaries. The trained model was integrated into a Tkinter-based Graphical User Interface (GUI) to visualize real-time detection outputs effectively. Implemented using Python-based tools, the framework is validated against benchmark lane and traffic signs, demonstrating better performance compared to existing systems in terms of lane detection, achieving 91.11% precision, and traffic sign detection, attaining 96.88% precision. Therefore, the proposed system delivers fast, accurate, and robust simultaneous lane and traffic sign detection, significantly improving real-time road safety and driver assistance.

Lane detection roadside traffic sign recognition contrast limited adaptive histogram equalization (CLAHE) self-artificial attentive gated neuro recurrent unit (SAAG-NRU) graphical user interface (GUI)

Abstract doi.org/10.14569/IJACSA.2026.0170843 PDF

44

HDST-CAF: Adaptive CNN Transformer Frequency Fusion for Deepfake Detection Under Visual Degradation

Author 1: Sanjitha S Laad Author 2: Smitha Shekar B

Digital media integrity is seriously threatened by deepfake technology, especially under real-world visual degradations such as poor resolution, noise, greyscale conversion, and compression artefacts. Unlike prior hybrid deepfake detectors that focus primarily on clean benchmark datasets, the proposed tri-stream framework is designed to improve robustness under multiple concurrent visual degradations through adaptive cross-attention fusion of spatial, semantic, and frequency-domain features. We propose HDST-CAF, a tri-stream architecture combining a lightweight Vision Transformer for global semantic inconsistencies, a multiscale CNN Feature Pyramid Network for local texture artefacts, and a Frequency Domain Analysis stream for GAN/diffusion fingerprints invisible in the spatial domain. The three streams are fused via bidirectional cross-attention with adaptive gating to dynamically weight complementary spatial and spectral cues conditioned on input degradation severity, trained with a joint main and auxiliary classification loss. On a custom multi-condition dataset of 506 videos (five degradation categories, approximately 5,060 extracted faces via YOLOv11), HDST-CAF achieves 92% overall accuracy, outperforming an identically evaluated ResNet18 baseline (89%) under most conditions, with 97% on greyscale and 96% on expression manipulation. Per-category accuracy ranges from 58% to 97%, with face swapping (74%) and low resolution (58%) remaining the most difficult cases. The proposed cross-attention tri-stream fusion offers a practical framework for deepfake forensics under real-world degradation, achieving a three-point improvement over the ResNet18 baseline in overall accuracy.

Vision transformer frequency domain analysis convolutional neural networks feature pyramid network multi-condition dataset cross-attention fusion and deepfake detection

Abstract doi.org/10.14569/IJACSA.2026.0170844 PDF

45

A Pipeline for Integrating and Analyzing Public Business Data in Albania Using Web Scraping and Machine Learning

Author 1: Markela Muca Author 2: Klodiana Bani Author 3: Amarildo Alla

The increasing availability of fragmented public business and procurement data creates opportunities for company-level empirical analysis, but limited evidence exists on how integrated procurement and administrative records can be used to characterize heterogeneous economic operators in Albania. This study investigates the analytical value of integrating public procurement records from the Public Procurement Agency (APP) with administrative business information from the National Business Center (QKB) through the unique NIPT identifier. The resulting company-level dataset is analyzed to examine whether procurement-derived characteristics contain systematic information for distinguishing heterogeneous company profiles and reproducing heuristic risk-oriented classifications. The empirical framework combines data normalization, entity-level integration, feature engineering, heuristic-label construction, supervised classification, Principal Component Analysis (PCA), K-Means clustering, anomaly detection, feature-importance analysis, and a financial-enrichment sensitivity experiment. In the 18-feature reduced experiment, HistGradientBoosting achieved the strongest mean repeated-cross-validation performance, with an F1-score of 0.8590, ROC AUC of 0.9683, and Average Precision (AP) of 0.9399. The analysis further identifies distinct company profiles characterized by differences in procurement intensity, contract value, and economic exposure, including small groups with exceptionally high activity and atypical characteristics. Financial enrichment did not produce a substantial improvement in classification performance, indicating that procurement-derived variables provide the dominant analytical signal in the current dataset. Across 999 deterministic label permutations, the mean ROC AUC was 0.4999, and the mean AP was 0.2700; the one-sided empirical p-value was 0.001 for both metrics. This diagnostic does not prove the absence of leakage, overfitting, or residual proxy dependence. The findings provide empirical evidence that company-level integration of procurement and administrative data can reveal heterogeneous procurement profiles and atypical patterns relevant to procurement monitoring, public-sector analytical capacity, and data-driven governance. Because the target labels are heuristic and are not independently validated against official risk outcomes, the results are interpreted as exploratory evidence for company profiling and risk-oriented analysis rather than as formal assessments of business risk, misconduct, or performance.

Web Scraping machine learning data integration public data analysis public procurement benchmarking Albania

Abstract doi.org/10.14569/IJACSA.2026.0170845 PDF

46

Revitalizing Cultural Heritage Education Through a Flow Theory-Based Serious Game Model for Enhancing User Engagement

Author 1: Wan Malini Wan Isa Author 2: Syadiah Nor Wan Shamsuddin Author 3: Norkhairani Abdul Rawi Author 4: Maizan Mat Amin Author 5: Nor Hafidzah Abdullah Author 6: Wan Mohd Adzim Wan Mohd Zain

Cultural heritage education plays a vital role in preserving national identity and transmitting traditional knowledge across generations. However, younger learners often perceive learning about cultural heritage as passive and less engaging than contemporary digital experiences. Although serious games have emerged as promising tools for heritage education, many existing applications lack theoretically grounded engagement mechanisms, limiting their effectiveness in sustaining learner participation. This study develops and empirically evaluates a Cultural Heritage Serious Game Model for Enhancing User Engagement Based on Flow Theory in the context of Songket learning. The research employed a four-phase methodology comprising analysis, model design, prototype development, and empirical evaluation. The proposed model integrates key Flow Theory-informed game design elements to support focused attention, intrinsic motivation, problem-solving, and learning transfer. Evaluation findings demonstrated positive engagement outcomes, with mean scores of 3.63 for focused attention, 3.98 for intrinsic motivation, 4.08 for problem-solving, and 4.02 for perceived learning transfer. The findings provide preliminary evidence of positive cognitive engagement in cultural heritage education. This study contributes a theoretically grounded design framework for serious game development and offers practical guidance for leveraging digital technologies to revitalize cultural heritage education.

Cultural heritage education serious games flow theory user engagement songket learning educational technology

Abstract doi.org/10.14569/IJACSA.2026.0170846 PDF

47

A Recursive Prompt-Driven Reasoning Approach for Multi-Hop Arabic Fatwa Question Answering

Author 1: Manal Al-Qahtani Author 2: Bader Alkhamees Author 3: Mourad Ykhlef

Answering Arabic fatwa inquiries represents a highly critical yet challenging task in Natural Language Processing, requiring consideration of multiple jurisprudential conditions, interpretation of interconnected concepts, and synthesis of evidence distributed across multiple religious sources. Existing Arabic question answering systems, including conventional single-step question answering (QA) models and retrieval-augmented generation (RAG) frameworks, typically perform retrieval only once and therefore struggle to capture the intermediate reasoning steps required for complex religious inquiries, often resulting in incomplete reasoning and factual hallucinations. To address these limitations, this study proposes a recursive prompt-driven reasoning approach for multi-hop Arabic fatwa question answering that integrates instruction-based supervision, task-specific language models, hybrid retrieval, and recursive reasoning to support end-to-end multi-hop question answering. The proposed method employs an instruction-based supervision strategy constructed from the MAFQA dataset to train specialized question decomposition and question answering models. To support evidence grounding, a hybrid retrieval module combines dense semantic retrieval, sparse lexical matching, and cross-encoder reranking to retrieve both few-shot demonstrations and supporting evidence at each reasoning hop, while a relevance-guided context update mechanism retains only the most informative contextual evidence throughout the recursive reasoning process. Comprehensive experiments conducted on the MAFQA dataset demonstrate the effectiveness of the proposed approach. The hybrid retrieval module consistently outperforms individual dense and sparse retrieval methods, while task-specific fine-tuning significantly improves end-to-end performance across Arabic sequence-to-sequence models and large language models under realistic inference conditions. The fine-tuned GPT-4.1 configuration achieves the best overall performance, obtaining an F1-token score of 31.88, a ROUGE-L score of 26.64, and a Relevance score of 92.39. Furthermore, the proposed approach surpasses representative state-of-the-art Arabic question answering systems, retrieval-augmented generation methods, and reasoning-based approaches, achieving a 36.5% relative improvement in F1-token over the strongest reasoning baseline. These results demonstrate that integrating recursive question decomposition, hybrid retrieval, and dynamic context updating improves performance under the evaluated multi-hop Arabic fatwa question answering setting.

Arabic question answering Arabic fatwa question answering Arabic natural language processing multi-hop reasoning retrieval-augmented generation hybrid retrieval large language models question decomposition

Abstract doi.org/10.14569/IJACSA.2026.0170847 PDF

48

Privacy-Aware Generative Augmentation for Imbalanced Intrusion Detection Using Differential Privacy

Author 1: Trung Ha Author 2: Tran Khanh Dang Author 3: Dinh Thi Hong Loan

Machine learning-based intrusion detection systems are often constrained by severe class imbalance, while generative augmentation may memorize distinctive network-flow records and expose membership information. This study evaluates privacy-aware augmentation for CICIDS2017 by integrating differentially private stochastic gradient descent (DP-SGD) into a conditional generative adversarial network and a variational autoencoder, yielding DP-CGAN and DP-VAE. The formal record-level privacy guarantee applies to the generative component and is tracked with an RDP accountant implemented through Opacus. Experiments use five target Benign-to-Attack augmentation ratios from one-to-one to one-to-five and five target privacy budgets of 0.5, 1, 2, 4, and 8. Random forest (RF) and XGBoost are evaluated using accuracy, Macro-F1, mean minority-class recall, and per-class recall for Bot, BruteForce, and Infiltration. Synthetic-data quality is examined through t-SNE, Wasserstein distance, mean squared error, and mean absolute error, while empirical privacy is assessed through a black-box membership-inference protocol using ROC-AUC, attack accuracy, and an accuracy-derived attack advantage. DP-CGAN preserves high downstream utility across both classifiers. DP-VAE remains competitive with random forest but degrades substantially with XGBoost, particularly for BruteForce and Infiltration. Membership-inference ROC-AUC remains close to 0.5 for both private and non-private generators under the implemented attack, indicating that no reliable membership signal was detected rather than proving the absence of leakage. These results show that private generative augmentation is feasible for imbalanced IDS learning, but its utility depends strongly on the interaction between generator architecture and downstream classifier.

Intrusion detection system class imbalance synthetic data generation differential privacy DP-SGD

Abstract doi.org/10.14569/IJACSA.2026.0170848 PDF

49

Leveraging Machine Learning for Predicting Marginal Field Oil Well Production Using Electrical Submersible Pump Operations Data

Author 1: Mohamed Zafir Bin Mohamad Gazali Author 2: Rabeea Jaffari Author 3: Areej Fatemah Meghji Author 4: Tuan Mohammad Yusoff Shah Author 5: Syed Muhammad Noaman Ahmed Shah Author 6: Muhammad Hammad Rasool

Electrical Submersible Pumps (ESPs) are critical components of oil well production that can be evaluated for their longevity through daily performance predictions. Traditional predictive methodologies have failed to handle the complexity of dynamic multi-well systems, creating a demand for automated Machine Learning (ML) pipelines based on operational data. In this study, we develop a multivariate time series ESP dataset based on factors impacting the total oil production of four wells (A1-A4) in the Jay field, Malaysia. A naive baseline pipeline used a 95%/5% chronological train-test split across pure regression models and a multivariate Prophet setup. However, due to the presence of data leakage, flat-trend autocorrelation, and the inability to process mechanical well shut-ins, the result of such an approach was unsatisfactory, producing extremely poor R² scores. In order to combat poor results and to find a more optimal solution, we have created a synchronized-lag multivariate pipeline evaluated across a 5-Fold Walk-Forward Cross-Validation framework. Noise has been eliminated by cutting leading zeros from the dataset pre-production, while stationarity has been provided through first-order differencing of target values for tracking daily changes (Δ y). To ensure temporal synchronization, a strict one-day chronological lag protocol has been used across all sixteen operational engineering features, splitting them up into a historical baseline (t-1) and an operational delta change feature (t-2 to t-1). The resulting input has allowed us to turn pure regression models into time series models forced to predict changes in targets based on purely historical pump parameters. Results show that non-linear ensemble models perform significantly better than Prophet under rolling validation. The Light Gradient Boosting Machine (LGBM) proved to be the champion model, achieving a statistically significant R² score of 0.31. Feature importance analysis showed that the operational adjustments of localized wells (specifically associated Gas_A2_delta and lag_oil_1) drive downstream fluid changes the most. Against the platform capacity of 9,500 barrels per day, LGBM's MAE translates to only 4.1% daily variation forecast.

Oil production prediction Electrical Submersible Pump (ESP) machine learning forecasting

Abstract doi.org/10.14569/IJACSA.2026.0170849 PDF

50

Explainable AI for Trustworthy 6G Edge-Cloud Smart Cities: A Systematic Review

Author 1: Riyad Almakki Author 2: Qaisar Abbas

Smart cities are becoming more interconnected, data-driven, and increasingly autonomous, with 6th-generation communication, edge-cloud computing, Internet of Things infrastructures, digital twins, federated learning, and artificial intelligence supporting them. The same technology that enables low-latency mobility, adaptive energy management, public safety analytics, environmental analytics, and responsive digital governance is creating a hard accountability conundrum. The decisions can be generated by black-box models across multiple computing tiers, take advantage of decentralized learning, and be executed before the human operator has time to review the reasoning. This review focuses on the trustworthy and explainable design of AI in these contexts. In this review, the focus is on the trustworthy and explainable design of AI in such contexts. The review is based on a multi-dimensional taxonomy of explanation timing, scope, target, representation, model dependence, computing location, stakeholder, trustworthiness objective, and urban application. It also outlines a tiered reference architecture that links local explanations at devices, operational explanations at mobile-edge nodes, network-level explanations for 6G orchestration, and strategic explanations for cloud-hosted urban digital twins. The synthesis confirms that the use of attribution techniques (SHAP, LIME, gradient-based saliency, feature importance) is prevalent, and that counterfactual explanations, concept-based explanations, causal explanations, inherently interpretable explanations, and interactive explanations are less developed for use in city-level deployments. Lacking are explanation fidelity for distribution shifts and real-time latency and energy cost, privacy leakage, federated explanation consistency, security against explanation manipulation, interoperability across urban digital twins, and weak human-centered evaluation. Finally, the study provides a research agenda and an engineering checklist of reliable, accountable, private, and sustainable urban intelligence.

Explainable artificial intelligence trustworthy artificial intelligence smart cities 6G networks edge intelligence cloud computing federated learning digital twins cybersecurity AI governance human-centered AI sustainable urban intelligence

Abstract doi.org/10.14569/IJACSA.2026.0170850 PDF

51

Feature Selection Versus Feature Extraction for IoT Intrusion Detection Under Temporal Evaluation

Author 1: Mahdia Abdessamad Author 2: El Bachir Tazi

Comparative studies of feature selection (FS) and feature extraction (FE) for IoT intrusion detection rely almost universally on random train/test splits, which let temporally adjacent, highly similar records appear on both sides of the partition. This study asks whether such conclusions survive when this evaluation leakage is removed. We reproduce the detection-performance component of the framework of Li et al. on the TON-IoT network dataset — identical preprocessing, correlation-based selection versus PCA-based extraction, five classifiers, and matched dimensionalities K ∈ {9, 22, 33, 47, 77} — varying only the partitioning rule across random stratified, global temporal, and per-class temporal protocols. At K = 9, the reference Decision Tree falls from 86.81% binary macro-F1 under random splitting to 58.04% under the global temporal protocol, in which unseen attack classes constitute 45.6% of the test records. In the multiclass task at K = 47, the same configuration declines from 93.14% to 14.26%, while the mean Matthews correlation coefficient (MCC) across 100 configurations falls from 0.638 to 0.196, a relative loss of approximately 69%. The family-level recommendation is directionally robust: for the single redundancy-oriented correlation selector evaluated here, PCA-based extraction outperforms selection in all three protocols (Wilcoxon over 50 matched pairs per protocol, p < 10⁻⁶). Configuration-level orderings and the identity of the best classifier, however, are protocol-dependent. Established for the specific selector and extractor evaluated, these findings indicate that feature-reduction guidance for IoT intrusion detection should be validated under leakage-resistant protocols, with macro-averaged metrics and MCC reported alongside accuracy.

Internet of things intrusion detection feature selection feature extraction data leakage temporal evaluation concept drift TON-IoT machine learning

Abstract doi.org/10.14569/IJACSA.2026.0170851 PDF

52

A Coordinate Transformation Framework for Vision-Guided Robotic Positioning in Automated Vehicle Body Marking

Author 1: Olzhas Olzhayev Author 2: Azhar Tursynova Author 3: Sayat Ibrayev

Vision-guided robotic positioning plays a critical role in modern manufacturing by enabling robots to accurately adapt to variations in workpiece position. This study proposes a coordinate transformation framework for automated vehicle body marking that converts visually detected target coordinates into executable robot motions through a unified processing pipeline. The proposed framework integrates visual target localization, homogeneous coordinate transformation, Tool Center Point (TCP) pose generation, and inverse kinematics to achieve accurate robotic positioning. Experimental validation was performed using a vision-guided robotic marking platform consisting of an AR4-MK3 six-degree-of-freedom manipulator, an overhead monocular camera, and a 50 W electromagnetic marking device. Five target positions with two repeated trials per position were used to evaluate positioning accuracy, repeatability, and computational performance. The proposed framework achieved an average positioning error of 2.19 mm, a maximum error of 2.69 mm, and a total processing time of approximately 27 ms, demonstrating stable and reliable robotic positioning during the marking process. The proposed framework provides a practical and computationally efficient solution for transforming visual information into robot-executable poses and offers a promising foundation for future vision-guided robotic marking systems operating in industrial environments.

Coordinate transformation vision-guided robotics robotic positioning vehicle body marking monocular vision tool center point (TCP) industrial robotics

Abstract doi.org/10.14569/IJACSA.2026.0170852 PDF

53

An Explainable Hierarchical Deep Representation Learning Framework for Multi-Class Host Rock Recognition from Mining Data

Author 1: Sami F Karali Author 2: Mohammed I Thanoon Author 3: Majed Nawas Author 4: Ahlem Fatnassi Author 5: Abed Saif Ahmed Alghawli Author 6: Zakhriya Alhassan Author 7: Yahia Said

This study aims to improve the automatic recognition of geological host rocks from mining data and to make the prediction process easier to understand for geologists, mining engineers, and data analysts. The study uses a mining dataset containing global mineral and rock information. The data are cleaned, missing values are handled, and the most useful features are selected through three feature selection methods. A deep learning model is then developed to learn hidden rock patterns, links among geological features, and differences between rock classes. The model combines several connected learning parts to capture local, regional, and broad geological patterns. A final classification step is used to identify different host rock types. Explanation tools are also applied to show which features most influence each prediction. The proposed model achieves strong results in recognizing several host rock classes, including limestone, andesite, diorite, gravel, granite, gneiss, sandstone, schist, quartzite, rhyolite, dolomite, alluvium, volcanic rock, and slate. The results show high accuracy in both training and testing data. The explanation results also help identify the geological features that contribute most to the model’s decisions. The experimental evaluation is conducted using the Mineral Ores Around the World dataset to assess the effectiveness of the proposed framework for host rock classification. However, the proposed model achieved strong classification performance. Under the 80:20 training-testing split, it obtained an accuracy of 97.91%, and under the 70:30 split, it achieved an accuracy of 97.31%. The study offers an explainable deep learning framework that combines feature selection, multi-level pattern learning, and prediction explanation for host rock recognition. The approach can support faster geological interpretation, mineral exploration, mining planning, and decision-making, and may guide future work on explainable artificial intelligence for mining applications.

Host rock recognition mining data feature selection explainable artificial intelligence geological feature selection mineral exploration multi-class rock recognition

Abstract doi.org/10.14569/IJACSA.2026.0170853 PDF

54

Effective Chunking for Retrieval-Augmented Generation over Structured Institutional Documents

Author 1: Siti Sarah Izhan Khalib Author 2: Abdul Hadi Abd Rahman Author 3: Lailatul Qadri Zakaria

Retrieval-Augmented Generation (RAG) has improved domain-specific question answering tasks by grounding responses in an external knowledge base. However, institutional documents differ from general corpora due to the presence of structured content such as tables, flowcharts, fee breakdowns and program codes. Existing RAG pipelines typically employ generic chunking strategies designed for unstructured text, which can fragment structured information and degrade answer quality. This study evaluates four chunking strategies across six configurations. The evaluation uses 65 publicly available structured documents from Universiti Kebangsaan Malaysia (UKM) with 28 Bahasa Melayu queries. Retrieval performance, answer quality and structural integrity are compared across configurations. Retrieval performance is generally similar across configurations, whereas answer quality varies more substantially. Structure-aware chunking achieves the highest ROUGE-L, BLEU and BERTScore F1 values, while semantic chunking achieves the highest table-integrity score. The findings demonstrate that chunking strategies influence RAG objectives in distinct ways, with structure-aware chunking improving answer generation while semantic chunking more effectively preserving table integrity. The results provide an empirical basis for selecting chunking configurations for structured institutional RAG systems.

Chunking retrieval-augmented generation structured text documents

Abstract doi.org/10.14569/IJACSA.2026.0170854 PDF

55

Explainable Neuro-Symbolic Learning for Mohs Hardness Prediction in Mining and Mineral Processing

Author 1: Mohammed I. Thanoon Author 2: Sami F. Karali Author 3: Mohamed H. Helal Author 4: Ahlem Fatnassi Author 5: Majed Nawas Author 6: Monir Abdullah Author 7: Figsma A. Alhashmi Author 8: Abdulbasit A. Darem

Predicting Mohs hardness from mineralogical information is still a challenging task because of the complex and nonlinear relationship between mineral composition and hardness properties. Traditional machine learning (ML) and deep learning (DL) strategies have been used to address this study, exploiting mineralogical attributes for prediction. Conversely, these techniques are frequently highly dependent on large labeled datasets, struggle to generalize over distinct mineral categories, and face challenges in extracting subtle physicochemical interactions and delivering interpretable predictions. This study presents an Explainable Deep Representation Learning Framework based on Mohs Hardness Prediction (XDRL-MHP). The primary objective of this work is to predict Mohs hardness using mineralogical properties by modeling mineral composition and structural characteristics. The proposed model employs mRMR for feature selection to identify the most informative physicochemical attributes. A physicochemical representation constructed using atomic interaction mapping, periodic table relationship encoding, chemical affinity representation, and molecular dependency construction to capture complex mineral relationships. Besides, a neuro-symbolic hybrid architecture is designed by integrating a capsule network for hierarchical mineral pattern learning, a neural tensor network for atomic relationship modeling, and a symbolic reasoning layer for rule-guided mineral intelligence. In addition, an explainable neuro-artificial intelligence technique incorporated to improve model interpretability and insight into predictions. The model is trained using the AdaBelief optimizer with quantile loss for ensure stable convergence and precise Mohs hardness prediction. The XDRL-MHP method validated utilizing the Comprehensive Database of Minerals, comprising 3,112 mineral instances with 140 mineralogical and physicochemical features. The simulation analysis of the XDRL-MHP model is performed using the Comprehensive Database of Minerals. Extensive comparative results show the effective performance of the XDRL-MHP model with an RMSE of 0.1140 over recent state-of-the-art models.

Mohs hardness prediction mineral mining adabelief optimizer explainable artificial intelligence physicochemical property

Abstract doi.org/10.14569/IJACSA.2026.0170855 PDF

56

Applying Visual Servoing for Industrial Robots Through Interactive Segmentation and Robot-Computer Communication

Author 1: Zeinel Momynkulov Author 2: Moldir Kuatova Author 3: Amandyk Tuleshov

Industrial robots are increasingly required to operate in flexible manufacturing environments where conventional offline programming and manually defined working positions limit production adaptability. This study presents a visual servoing framework that integrates interactive image segmentation with industrial robot control through robot-computer communication. The proposed system enables an operator to specify the desired manipulation region directly on an RGB-D camera image, eliminating the need for predefined object models or manually generated trajectories. A fine-tuned Segment Anything Model (SAM) is employed to obtain accurate segmentation masks under industrial conditions, while depth information and hand-eye calibration are used to transform the selected image region into robot coordinates. The generated positions are transmitted to an industrial robot through a TCP/IP communication interface, where they are executed by the robot controller. A graphical user interface was developed to combine image acquisition, interactive segmentation, trajectory visualization, and communication monitoring within a single application. Experimental evaluation demonstrated reliable robot-computer communication and stable segmentation performance in the presence of illumination changes, metallic reflections, and partial occlusions. The proposed segmentation module achieved an average confidence of approximately 91% with an average inference time of 46 ms, enabling near real-time operation for industrial visual servoing. The obtained results indicate that the proposed framework simplifies robot deployment while maintaining accurate target localization and compatibility with existing industrial robotic platforms, making it a practical solution for flexible manufacturing applications.

Industrial robotics interactive segmentation human-robot interaction robot-computer communication industrial automation

Abstract doi.org/10.14569/IJACSA.2026.0170856 PDF

57

SHAP Guided Transformer-Based Deep Learning Model for Diamond Clarity Grading in Smart Mining and Resource Valuation Systems

Author 1: Ahlem Fatnassi Author 2: Shouki A. Ebad Author 3: Sami F Karali Author 4: Mohammed I Thanoon Author 5: Majed Nawas Author 6: Monir Abdullah Author 7: Asma A. Alhashmi Author 8: Abdulbasit A. Darem

Diamond clarity grading aims to automatically determine the quality of diamonds by analyzing visible inclusions and structural flaws using intelligent analysis methods. By integrating diamond cost prediction with clarity grading, a unified architecture can be employed for both quality examination and market value evaluation, where clarity attributes support the determination of the diamond's economic value. Traditional machine learning (ML) algorithms depend upon handcrafted characteristics that frequently miss composite structural and visual patterns. Deep learning (DL) enhances feature learning; however, it faces difficulties such as data imbalance, limited interpretability, and inadequate generalization. Consequently, the real-time diamond grading outcome remains limited. Therefore, this study proposes a SHAP-Guided Transformer Network (SGT-Net) for diamond clarity grading. The proposed model includes a preprocessing stage that prepares the diamond grading dataset for effective and reliable model training. Moreover, feature engineering is carried out to enhance the representation of the dataset by deriving meaningful attributes from the raw diamond features. Moreover, SHAP-guided feature prioritization is applied to identify the most influential features in the data. The transformer-based architecture processes the data using separate embeddings for numerical and categorical features, which are then passed through a transformer encoder with multi-head self-attention. In this setup, SHAP is used to understand feature importance and guide the model to pay more attention to the most relevant features during learning. The resulting fused feature representations are then passed through a dense layer followed by a softmax classifier to predict diamond clarity into 8 distinct classes. For training, the model is optimized using the RAdam optimizer along with a weighted cross-entropy loss function. Experimental evaluation is conducted using the Diamond dataset with 8 clarity classes and 53940 samples from the Kaggle repository. The results demonstrate that the proposed SGT-Net effectively learns complex feature relationships and improves diamond clarity grading performance with enhanced accuracy of 96.59%, making it suitable for resource valuation systems.

Diamond clarity grading quality assessment deep learning multi-head self-attention model optimization

Abstract doi.org/10.14569/IJACSA.2026.0170857 PDF

58

Classification of Lymph Node Tumor Images Using Adaptive Filtering in Convolutional Neural Networks

Author 1: Jamal M. Alothman Author 2: Nameer N. El. Emam Author 3: Issa Qabajeh Author 4: Mhammed Almakadmeh

Tumors of the lymphatic system can be either benign or malig-nant. However, because traditional diagnostic approaches rely on subjective assessment by individual pathologists and often lead to significant discrepancies and misclassifications, early and accurate diagnosis is important for successful treatment and improved scientific outcomes. We compare an adaptive convo-lutional neural network (CNN) model, trained on filter shapes, with regular CNNs by benchmarking them for image classifica-tion. This adaptive model automatically retunes filter size based on statistical tests of images, improving feature extraction and classification performance. More specifically, it achieves an accuracy of around 92%, which is about 10- 13% higher than random predictions using most standard models; however, it also has high precision and recall, 94% and approximately 4.5% higher in this metric, and 85%, with a further 27% in-crease relative to standard benchmarks used for others. An F1-score of 89% represents an 18% uplift, while AUC, at 97, outperforms the same model without ours by up to 11%. In ad-dition, the mean squared error (MSE) is reduced to 0.096 for the adaptive model, compared with 0.24 for traditional methods. The results show that an adaptive filter can benefit CNN-based image classification. This exploratory research opens a new opportunity to apply deep learning methodologies in practice, supporting their integration into real-world applications such as medical imaging, autonomous systems, and security analysis. Although a small step, it shows the model as an efficient, fast, and scalable classification solution.

Convolutional neural networks adaptive image filtering lymph node tumor images image classification feature extraction

Abstract doi.org/10.14569/IJACSA.2026.0170858 PDF

59

An Early Warning System for Proactive DDoS Attack Detection Using Ordinary Differential Equation-Based Scoring

Author 1: Radhakrishna Vangipuram Author 2: Rajput Eswar Sai Singh Author 3: Ramesh Kumar Aytha Author 4: Sravankiran Vangipuram Author 5: Sreenivasa Rao Annaluri

Distributed Denial of Service (DDoS) attacks are still a significant issue facing cybersecurity, flooding infrastructure with large packet torrents and disrupting networks across the globe. To mitigate against such volumetric attacks, the best approach is to intercept the threat long before any attack gets underway. However, conventional defense systems react primarily after the volume of attack peaks; thus, they leave critical assets vulnerable to disruption. To avoid this, it is essential to detect and analyze precursor signals within the attack preparation period. This work introduces a framework for generating an early warning signal based on computed scores and set thresholds to categorize network traffic alerts into low-, medium-, and high-impact alerts. In this study, a differential equation model is applied to the CICDDoS2019 dataset to generate these alert scores. The goal is to make a prediction well before the attack blows up in size and before the onset of the attack. The results exhibit an impressive capacity for proactive detection, from reactive mitigation to predictive defense. This framework was able to predict the intrusion with low, medium, and high alerts issued 46 minutes and 53 seconds, 46 minutes and 49 seconds, and 46 minutes and 55 seconds before the attack peak. The proposed framework obtained a balanced accuracy ranging from 99.43% to 99.70% for these three alert classifications.

DDoS early warning signals ordinary differential equation residual low threshold medium threshold high threshold

Abstract doi.org/10.14569/IJACSA.2026.0170859 PDF

60

Semantic-Driven Community Detection in Complex Networks Using Fuzzy Logic and Multi-Criteria Decision Making

Author 1: Mohamed El-Moussaoui Author 2: Mohamed Hanine Author 3: Sulieman S. Alshuhri Author 4: Amal Alomran Author 5: Amir Mohamed Talib

The study introduces a novel approach to community detection in complex networks by integrating fuzzy logic with multi-criteria decision-making techniques. Unlike traditional methods that rely primarily on topological metrics, the proposed approach incorporates semantic attributes to identify meaningful community structures. Fuzzy logic addresses the inherent uncertainty and ambiguity in processing these attributes, enabling a flexible detection process that is not dependent on network topology. To enhance scalability, this study customizes the k-means clustering algorithm to accommodate small- and large-scale network structures. Experimental results show that the proposed fuzzy logic-based approach achieves competitive performance compared with conventional algorithms. Additionally, the proposed approach demonstrates robustness by generating well-balanced communities with competitive execution times compared with the evaluated methods. These findings highlight the potential benefits of incorporating semantic attributes and fuzzy reasoning into community detection in complex networks.

Community detection fuzzy logic Multi-Criteria Decision Making (MCDM) social network analysis semantic communities

Abstract doi.org/10.14569/IJACSA.2026.0170860 PDF

61

Influence of Big Data Analytics Capability on Employee Creative Behavior: A Modified Mediation Model

Author 1: Mohammed Bassam Nassoura Author 2: Nabil Abudarwish Zarqa Author 3: Mohammad Musa Al-Momani Author 4: Bilal Alnassar Author 5: AbdelRahman Ismail Author 6: Ahmad Awadallah

The purpose of the present study was to assess the effect of Big Data Analytics capabilities on employees' creative behavior in the Jordanian service sector and to explore the mediating influence of employee curiosity and the moderating influence of top management support. The proposed model was developed using an integrative theoretical framework, namely the knowledge-based view (KBV) and Self-Determination Theory (SDT), to understand how knowledge resources translate into creative behaviors through psychological motivations and organizational context. Data for this quantitative research were collected from a population of 326 employees working in SMEs operating in the Jordanian service sector. Statistical analysis of the collected data was performed using Partial Least Squares Structural Equation Modeling (PLS-SEM). The findings have confirmed the direct positive influence of big data analytics capabilities on creative behavior, alongside an indirect one via employee curiosity as a mediating factor. It was also found that senior management support has a negative moderating effect on the relationship between big data analytics capabilities and employee curiosity. This research study adds value to the body of knowledge by combining two theoretical frameworks that relate cognitive resources, psychological motivations, and organization in the explanation of creativity in the digital business context.

Big data analytics employee creative behavior curiosity management support the knowledge-based view and self-determination theory

Abstract doi.org/10.14569/IJACSA.2026.0170861 PDF

62

GTA: Automated Semantic NLP-Based Generation of UML Activity Diagrams from Acceptance Criteria

Author 1: Samia Nasiri Author 2: Salim Bloundi Author 3: Mohammed Lahmer

In agile development, user stories express stakeholder needs, and the associated acceptance criteria (AC), written in the Given/When/Then (GWT) notation, specify the behaviour expected of the system under stated preconditions. Their manual translation into UML activity diagrams is laborious and remains sensitive to wording differences between authors. Existing automated approaches rely on lexical matching or on grammatical rules. Both fail when an identical precondition is expressed through different wording, a case that occurs frequently in collaborative backlogs. This study introduces GTA (GWT-to- Activity), a deterministic pipeline that converts GWT AC into UML activity diagrams. The clause structure is first recovered through dependency parsing. Paraphrased Given states are then merged based on sentence embedding similarity using Sentence-BERT (SBERT), and natural language inference (NLI) is used to separate semantic opposition from lexical variation, with thresholds calibrated on annotated data to preserve output stability under paraphrastic variation. Eleven behavioural patterns are supported, from simple sequences to nested compound conditions. The pipeline was evaluated on 107 test sets containing 230 AC drawn from four heterogeneous corpora of synthetic, open-source, and benchmark origin, and it reaches a node F1-score of 99.26%, an edge F1-score of 97.37%, and an exact topological match rate of 95.33%. All eleven patterns are recovered, and every generated file is a valid PlantUML that requires no post-processing. The output is structurally correct, deterministic from one run to another, and traceable to the thresholds set at each gate.

Behaviour-driven development GWT acceptance criteria UML activity diagrams sentence-BERT NLI natural language processing requirements engineering

Abstract doi.org/10.14569/IJACSA.2026.0170862 PDF

63

A Novel Multi-Objective Service Scheduling Framework for Fog-Driven Agricultural Internet of Things

Author 1: Mengya NIE Author 2: Fengmin JIAO

Timely, context-aware service recommendations are most significant for maximizing system efficiency and resource utilization in fog-enabled Internet of Things (IoT) for smart agriculture. Classical Grey Wolf Optimization (GWO) algorithms are prone to premature convergence and lack robustness for multi-objective and dynamic environments. This study presents an Enhanced Grey Wolf Optimization (EGWO) algorithm that incorporates adaptive weighting and nonlinear exploration-exploitation control. The EGWO algorithm uses a dynamic alpha-beta-delta hierarchy for weighting and a cosine-based decay for the control parameter, ensuring better convergence and global search efficiency. In the fog-based service recommendation environment, EGWO optimizes multiple conflicting objectives, including latency, energy consumption, trust, and service utility. A comparison is made between standard GWO, PSO, and DE using a simulated agricultural fog-IoT network. The results demonstrate that EGWO achieves faster convergence speed, higher recommendation quality, and uniform behavior across all test scenarios. The method is suitable for practical application in real-time agricultural use cases, ensuring guaranteed optimal service allocation and improved decision support at the network's edge.

Grey wolf fog computing internet of things smart agriculture service recommendation

Abstract doi.org/10.14569/IJACSA.2026.0170863 PDF

64

CTU-Tutor: Contextual and User-Aware Language Understanding for Personalized Tutoring with Long-Form Text Generation

Author 1: Akanksha Bisht Author 2: Oshin Sharma

This study introduces CTU-Tutor, a contextual and user-aware intelligent tutoring system for generating personalized long-form text as learning content. The system combines BERT-based learner embeddings, K-means for proficiency clustering, Transformer-XL for long-context modelling, BiLSTM-CRF for concept extraction, and LDA-based knowledge graph construction. A retrieval-augmented Transformer-XL model produces personalized content, which is supplemented with explainable AI to make decisions transparently using SHAP and LIME. Experimental analysis shows that CTU-Tutor is better than state-of-the-art models, such as ExPerT, REST-PG, GSPT-CVAE, LONGLaMP, and Transformer_QA, in various metrics. The proposed framework has 0.98 accuracy, 0.98 F1-score, 0.97 MCC, and 0.98 precision, sensitivity, and specificity, and minimized false negatives (0.013) and false positives (0.017). For text generation, BLEU, ROUGE, and METEOR scores were approximately 0.98, indicating high overlap with the corresponding human-authored reference texts. These results prove that the synergistic implementation of learner profiling, long-context modelling, structured knowledge representation, and explainable AI delivers more precise, reliable, and pedagogically transparent personalized tutoring.

Contextual and user-aware tutoring system bidirectional encoder representations of transformers transformer-XL bidirectional long short-term memory with conditional random fields explainable AI latent dirichlet allocation

Abstract doi.org/10.14569/IJACSA.2026.0170864 PDF

65

Cryptographic Attestation Against Integrity Attacks in Service Monitoring: A Threat Model and Verifiable Architecture

Author 1: M A Anusuya Author 2: Chayadevi M L Author 3: Shraddha C Author 4: Vani H Y

Service uptime monitoring infrastructure is a high-value target for data-integrity attacks: a single compromised or dishonest monitoring provider can fabricate availability records, retroactively suppress outage evidence, or silently alter historical data, and clients today have no cryptographic means of detecting such manipulation. This study develops a threat model for monitoring-data integrity attacks—covering provider-side tampering, evidence suppression, Sybil-identity flooding, and submission replay—and presents a verification architecture engineered to resist each threat in that model. Independent validator nodes sign availability observations with Ed25519 keys; a quorum-based aggregation rule tolerates up to f < Q/2 Byzantine validators, SHA-256 content-hash commitments bind off-chain evidence to an immutable on-ledger record that any third party can independently re-derive and check without trusting the aggregator; and stake-bonded registration imposes a quantifiable capital cost on Sybil identities. We formalize the adversary model, prove signature unforgeability under the Elliptic Curve Discrete Logarithm assumption, derive the capital cost of quorum capture, and bound the residual attack surface—selective evidence inclusion and round-stalling—that persists even under a semi-honest aggregator. A seven-day, five-validator, three-region deployment achieves 99.7% quorum agreement, sub-6-second worst-case attestation latency, and zero false positives or negatives across 200 independently re-verified historical rounds, confirming that the architecture removes the central point of trust that lets a single compromised provider corrupt monitoring evidence undetected.

Cryptographic attestation threat modelling Byzantine fault tolerance Sybil resistance digital signatures data integrity tamper-evident logging adversarial security analysis non-repudiation

Abstract doi.org/10.14569/IJACSA.2026.0170865 PDF

66

Territorial Governance in Transition:Multi-Level Frameworks, Citizen Participation,and Digital Transformation (2018–2026)

Author 1: Abdelhakim Errakha Author 2: Sanaa DFOUF Author 3: Daafi Redouan Author 4: Fekkak HAMDI

From 2018 to 2025, the combination of the democratic deficit and growing digitalization caused significant changes to territorial governance. To understand these changes, this study assembles the data of more than thirty countries from five different continents and describes the following three interrelated phenomena: the restructuring of multi-level governance systems, the establishment of participatory practices at the subnational level, and the growing use of digital instruments to transform public accountability and service provision. Using a longitudinal framework, the study develops a composite governance frame-work and provides several case studies. The study identifies a path-dependent, convergent model of decentralization, whereby governance systems brought participatory practices closer to citizens and created new challenges for governance systems integration. The results of the study indicate the first tangible benefits of advanced territorial governance in the regions of interest. The results of the study also show that from 2018 to 2025, the countries of the study witnessed significant progress in decentralization (the level of decentralization increased in all benchmark countries), and there was a significant improvement in the Citizen Participation Index: European Union (from 58 to 72), Latin America (from 45 to 60), Southeast Asia (from 38 to 55), and Sub-Saharan Africa (from 32 to 50). When digital platforms are incorporated into strong institutional systems, they improve transparency, citizen participation, and service improvement. Nevertheless, unequal digital access and imbalanced regional administrative capacity result in differing regional systems of territorial governance. The study also provides an integrated policy system for the digitally integrated, inclusive, resilient, and adaptive territorial governance systems and the governance challenges posed by the COVID-19 pandemic and the Sustainable Development Goals.

Territorial governance decentralisation multi-level governance citizen participation digital transformation regional development public policy

Abstract doi.org/10.14569/IJACSA.2026.0170866 PDF

67

A Feature Model-Based Configurable Framework for Reproducible Intervention Evaluation

Author 1: Carlos Murillo-Barrera Author 2: Miguel Vera Author 3: Nicolás Márquez Author 4: Cristian Vidal-Silva

Intervention evaluation frequently relies on fixed analytical workflows that provide limited support for methodological adaptability, reproducibility, and systematic management of analytical variability. This study proposes a Feature Model-based configurable framework that represents data preparation, feature construction, statistical inference, robustness validation, and decision-support generation as interdependent analytical features. Mandatory, optional, alternative, and dependency-constrained decisions are modeled explicitly, enabling the systematic generation, validation, and diagnosis of reproducible analytical configurations. The framework was instantiated in a managerial capability development intervention involving micro-enterprises. Across the strategic, financial, and operational capability dimensions, the generated pipelines identified statistically significant improvements with Cohen’s d values ranging from 0.48 to 0.65. The direction and inferential interpretation of the intervention effects remained stable across the admissible preprocessing, imputation, and testing configurations evaluated in the case study. Because the empirical assessment uses a single dataset from one application domain, the findings establish technical feasibility and within-case robustness rather than cross-domain generalizability. The study contributes a variability-aware analytical architecture and an open-source Python-based engine integrating Feature Model reasoning, configuration validation, and Parallel FastDiag-based conflict diagnosis, thereby supporting transparent, inspectable, and adaptable intervention-evaluation workflows.

Feature models variability management configurable systems reproducible analytics decision support systems intervention evaluation

Abstract doi.org/10.14569/IJACSA.2026.0170867 PDF

68

A New Metadata-Aware Retrieval-Augmented Generation (RAG) Architecture for Trustworthy Legal Question Answering

Author 1: Alexandra V. Jove-Ticona Author 2: Luis J. Duarte-Coaquera Author 3: Israel N. Chaparro-Cruz Author 4: Silvana B. Cabana-Yupanqui Author 5: Americo Chaparro-Guerra

Large Language Models (LLMs) offer strong capabilities for Natural Language Processing, yet their inherent uncertainty often produces hallucinations, confident but incorrect statements, which is critical in domains requiring precise knowledge representation. Retrieval-Augmented Generation (RAG) reduces this risk through information retrieval, but standard pipelines still suffer from fragmented context and weak alignment between queries and legal provisions, limiting trustworthy knowledge extraction. This study proposes a Metadata-Aware RAG architecture to improve grounding in large legal corpora. It integrates: 1) Sub-chunking with Legal Metadata Inheritance, which transforms unstructured legal PDFs into granular, metadata-rich fragments; and 2) an Adaptive Filter Creator, a pipeline that extracts structured constraints and compiles optimized hybrid retrieval queries. These components enhance semantic alignment, reduce uncertainty-driven hallucinations, and strengthen neural information retrieval. Using a curated Peruvian labor law corpus and 150 manually validated question–answer pairs, the system was evaluated across three LLMs (Llama-3.1-8B, GPT-OSS-20B, Gemma-3-27B). The proposed architecture achieves double-digit improvements over a Naive RAG baseline across all four RAGAS metrics—Context Precision, Context Recall, Factual Correctness, and Faithfulness—with gains ranging from 13.10% to 28.15%; notably, Faithfulness surpasses 0.90 for Gemma-3-27B. Statistical analysis confirms significance (t(11) = 15.49, p = 4.06 × 10−9) with an extremely large effect size (Cohen’s d = 4.47). Regression results show minimal influence of model size (slope < 0.005), indicating that retrieval design has a stronger influence than parameter count in the evaluated setting.

RAG Retrieval-Augmented Generation LLMs legal question answering

Abstract doi.org/10.14569/IJACSA.2026.0170868 PDF

69

AlphaOCR: Integrated Deep Learning and Optical Character Recognition for Receipt Extraction

Author 1: Muhammad Haikal Iman Osman Author 2: Nor Samsiah Sani Author 3: Luan Xiang Wei Liu Author 4: Zalinda Othman Author 5: Mohd Aliff Afira Sani Author 6: Ibrahim Zebiri

Receipts are vital documents that validate transactions by capturing key details such as dates, items purchased, prices, and seller information. However, accurately documenting receipts is often compromised by issues like blurring, which can result from poor image quality, physical wear, or suboptimal scanning conditions. While the Processing Key Information Extraction from Documents Using Improved Graph Learning-Convolutional Networks (PICK) deep learning model is effective for structured information extraction, it struggles with processing blurred text, leading to inaccuracies in data retrieval. To overcome these challenges, this research introduces AlphaOCR, a web-based application that integrates the PICK model with advanced Op-tical Character Recognition (OCR) technology. The methodology integrates OCR-based line-item recognition with PICK-based structured-field extraction to extend the range of information recovered from receipts. The deep learning component was evaluated using mean Entity Precision (mEP), mean Entity Recall (mER), mean Entity F1-score (mEF), and mean Entity Accuracy (mEA), together with a paired t-test for the entity-level mEF results. PICK-2 increased the mEF from 83.32% to 83.88%, corresponding to a modest absolute gain of 0.56 percentage points. The complete AlphaOCR configuration combining PICK- 2 with EasyOCR achieved an overall accuracy of 87.96% under the evaluation setting used in this study. Among the OCR models evaluated, EasyOCR achieved a Character Accuracy Rate (CAR) of 92.04% and provided the most suitable output characteristics for integration with PICK-2 within the tested configuration. The reported results should be interpreted within the experimental scope of this study rather than as a direct performance ranking against previously published systems that used different datasets, tasks, extracted fields, and evaluation metrics. AlphaOCR provides a practical integrated workflow for receipt information extraction, although broader claims regarding robustness, scalability, and generalisability require evaluation on larger and more diverse receipt collections. This research contributes to receipt information extraction primarily through the system-level integration of graph-based key information extraction, OCR-based item recognition, output fusion, and a web-based verification workflow. Future work should evaluate the system on larger and more diverse datasets and investigate enhanced semantic understanding and additional document types.

AlphaOCR optical character recognition information extraction receipt processing deep learning

Abstract doi.org/10.14569/IJACSA.2026.0170869 PDF

70

Federated Defense in the Medical Edge: A Reputation-Aware Secure Aggregation Protocol for IoMT Intrusion Detection

Author 1: Feras Fares AL-Mashakbah Author 2: Abdullah Alqammaz Author 3: Ala’ Khalifeh Author 4: Eman Fares Al Mashagbah Author 5: Essam Said Hanandeh

Smart hospitals deploy Internet of Medical Things (IoMT) sensors and MQTT brokers to stream clinical telemetry over resource-constrained edge gateways. Centralized network intrusion detection systems (NIDS) expose sensitive traffic traces and create single-point failures; federated learning (FL) avoids raw-data centralization but remains vulnerable to client-side model poisoning and server-side inspection of client updates. This study presents Trust-Weighted Federated Aggregation (TW-Fed), a reputation-aware defense framework for federated learning that integrates 1) cosine similarity-based trust scoring using low-dimensional update sketches and 2) secure aggregation of trust- and data size-weighted model updates through pairwise masking with dropout recovery. TW-Fed has been evaluated on the CICIoMT2024 benchmark (Wi-Fi and MQTT subsets) with a four-class NIDS task (Benign, MQTT-Connect-Flood, TCP-DDoS, ARP-Spoofing) and 20 edge clients emulating IoMT gateways. In the clean setting, TW-Fed achieves 99.21% accuracy and 98.74% macro-F1, exceeding FedAvg by 0.82 macro-F1 points. Under label flipping with 30% malicious clients, TW-Fed sustains 97.06% accuracy versus 82.14% (FedAvg), 85.33% (FedProx), and 92.07% (Krum). Under backdoor injection, TW-Fed reduces attack success rate from 91.5% (FedAvg) to 6.8%. On Raspberry Pi 4 clients, TW-Fed adds 0.09 s local overhead per round while reducing server aggregation time by 34% relative to Krum. Secure aggregation increases uplink volume by 9.7% while preventing per-client update disclosure.

Federated learning intrusion detection internet of medical things secure aggregation poisoning attacks reputation systems MQTT security

Abstract doi.org/10.14569/IJACSA.2026.0170870 PDF

71

An Initial Validation of a Pedagogical Simplification Gain Index for Arabic Technical Text Simplification

Author 1: Khalid Essaadani Author 2: Soumia Ziti Author 3: Karima Salah-Eddine

Simply calculating the readability score or the degree of lexical overlap for Arabic technical text simplification does not provide the full picture because an explanation that is shorter may still damage the technical meaning, weaken terminology consistency, or fail to reduce learner difficulty. This study introduces the Pedagogical Simplification Gain Index (PSGI Auto), which is automatic and interpretable, and is designed to measure the pedagogical gain brought about by simplification. Three elements are integrated into PSGI Auto: meaning preservation gain, terminology clarity gain, and concision gain. The technique was tested on Arabic simplification outputs sourced from five information technology areas, which provided a human reference score obtained from three independent expert raters. There is a strong and statistically significant correlation between PSGI Auto and human PSGI judgments, with a Pearson correlation of 0.779, Spearman correlation of 0.755, and 81.6% direction agreement. The results suggest that PSGI Auto can possibly be used as a practical and interpretable learner-oriented Arabic technical simplification tool that would help to identify whether simplification results in pedagogical improvement, not just a superficial change in the text. Since the validation is carried out on a relatively small dataset, the metric is to be interpreted as an initial one requiring further validation by comparison with well-established simplification metrics and larger-scale testing.

Arabic technical text simplification automatic text simplification pedagogical evaluation PSGI Auto human evaluation

Abstract doi.org/10.14569/IJACSA.2026.0170871 PDF

72

Sustainability-Centric Software Development: A Quantitative Framework for the SDLC

Author 1: Madhura G K Author 2: Piyush Kumar Pareek

Although software systems increasingly shape energy consumption, economic output, and societal welfare, most development methods still prioritise schedule, cost, and functionality. This paper presents an approach to software development that prioritises sustainability by embedding environmental, economic, and social objectives into the SDLC from the very beginning. The framework represents sustainability as a set of quantifiable variables that are combined into a Global Sustainability Index (GSI). These metrics include operational energy and carbon footprint, total cost of ownership (TCO), maintainability index, defect density, and a normalised Social Impact Score (SIS). An empirical measurement architecture gathers runtime and process data for continuous improvement, while phase-level “sustainability budgets” direct trade-offs across requirements, design, implementation, testing, and operation. The framework is evaluated in a repeated-measures industrial study of six production software systems (758 KLOC in total, 58 engineers, six application domains), in which every system is observed over four counterbalanced release cycles governed respectively by the proposed framework and by three established approaches: GREENSOFT, GreenSDLC, and the Sustainability Quality Model (SQM). The proposed framework attains the highest GSI (0.86 ± 0.03), a statistically significant improvement of 10–19% over the competing frameworks (paired t-tests, all Holm-adjusted p < 0.002, Cohen’s dz > 2.5). Relative to current models, energy usage and carbon emissions are cut by 10–20%, and they are decreased by approximately 30% when compared to a no-framework baseline. Normalised maintainability and defect density both improve over a five-year timeframe, and total cost of ownership drops 4–9%. Consistently higher levels of social impact and stakeholder satisfaction are observed, particularly for user groups who are marginalised. Ninety-five percent confidence intervals and effect sizes are reported for every headline comparison, and the principal limitations of the framework are stated explicitly together with mitigation strategies. These results show that all three dimensions can be improved with explicit quantitative sustainability integration without a rise in long-term costs.

Sustainability-centric framework social impact score total cost of ownership global sustainability index software development lifecycle

Abstract doi.org/10.14569/IJACSA.2026.0170872 PDF

73

Definitional Label Leakage in Aquaculture Water Quality Anomaly Detection

Author 1: Rosida Vivin Nahari Author 2: Teguh Prasetyo Author 3: Riza Alfita

Semi-supervised anomaly detection is widely pro-posed for Internet of Things water quality monitoring in aquaculture. However, reported high scores often result from definitional label leakage. This occurs when an observation is labelled anomalous based on a physicochemical threshold, and that same variable is used as an input feature. The benchmark then measures threshold recovery rather than true anomaly detection. To quantify this artefact, we introduce two trivial reference detectors: an all-positive predictor and a per-channel range rule. Using a rigorous leakage-aware protocol, we evaluated Isolation Forest, one-class SVM, a Liquid State Machine, a convolutional autoencoder, and an LSTM autoencoder on two corpora: the three ponds of a public aquaponics dataset that survive a sensor-integrity screen fixed before modelling, eight of eleven ponds having been rejected, and a six-pond dataset from East Java in which all positive test windows originate from two ponds. One-class SVM proved the strongest detector on both corpora, achieving a Matthews correlation coefficient up to 0.77, although with two corpora the shared detector ordering is indicative rather than established. Crucially, on the public corpus, both autoencoders and the trivial range rule performed identically poorly, with recall capped near twenty-one percent. This collapse stems from normal-class contamination caused by duration-based labelling, which inflates the percentile threshold for reconstruction-based models. Furthermore, we demonstrate that the F1 score of a zero-information all-positive predictor exceeds several trained detectors, rendering F1 unsafe as a primary metric at these prevalences. Because the labels are derived from husbandry thresholds rather than expert annotation, and because our multi-seed statistics quantify reproducibility on fixed partitions rather than generalisation across farms or seasons, these results bound the interpretation of the benchmark and not field performance. We conclude by proposing a strict reporting protocol for aquaculture anomaly detection studies, with guidance for applying it retroactively to already-published corpora.

Anomaly detection aquaculture water quality label leakage one-class SVM liquid state machine autoencoder

Abstract doi.org/10.14569/IJACSA.2026.0170873 PDF

74

Hybrid Blockchain-Based Tokenization for Secure and Interoperable Electronic Payment Systems

Author 1: Omniah Abdullah Al Ibrahim Author 2: Suhair Alshehri

The rapid evolution of digital payment technologies has highlighted critical challenges in ensuring data security, transparency, and compliance across centralized tokenization systems. Traditional models such as those used by EMVCo and major card networks rely on Token Service Providers (TSPs) and token vaults, which, while effective at masking Primary Account Numbers (PANs), still suffer from single points of failure, limited auditability, and potential privacy risks. Fully decentralized blockchain-based payment models, on the other hand, often face latency, scalability, and regulatory compliance challenges that hinder real-world adoption. This paper proposes a blockchain-based distributed tokenization model that decentralizes the tokenization and validation phases while maintaining centralized authorization and settlement through existing ISO 8583 payment rails. Unlike traditional centralized tokenization architectures that rely on a central TSP for token management, and fully decentralized approaches that migrate payment processing onto the blockchain, the proposed model selectively decentralizes token lifecycle functions while preserving compatibility with existing payment infrastructure. The model introduces a two-layer architecture: an on-chain layer for token creation, validation, and consumption using smart contracts deployed on the Ethereum network, and an off-chain layer for detokenization, authorization, and regulatory compliance through the issuer-side Token resolution service (TRS) and Token Vault. Experimental results show a 65–70% reduction in gas consumption and lower transaction latency compared to a fully on-chain reference model. These improvements stem from the model’s hybrid structure, which minimizes state changes and optimizes network resource utilization. The findings indicate that the proposed model enhances security, interoperability, and operational efficiency while supporting regulatory compliance and blockchain-based token lifecycle management within real-time payment workflows, while keeping latency-critical authorization and settlement off-chain.

Blockchain-based payment systems decentralized tokenization electronic payment security smart contracts hybrid payment model

Abstract doi.org/10.14569/IJACSA.2026.0170874 PDF

75

Effectiveness of Adaptive Shape Prediction Model for Cytological Cell Segmentation

Author 1: Afaf Omar Tareef

Cell segmentation continues to be a significant challenge in medical image processing, as traditional segmentation methods fail to provide precise and thorough segmentation of complex structures such as touching and overlapping cells. The geometric data points of these cells are inadequate for accurate contour estimation to yield a correct segmentation. Adaptive shape priors have demonstrated efficacy as a promising solution in these challenging circumstances. They can be employed to restrict segmentation models in medical imaging and enhance segmentation robustness in the presence of a complex structure of touched and overlapped cells. In this study, Adaptive Shape Prior Model (ASPM) is introduced and evaluated using three cervical cell datasets, where all their images have a varied number of cells with different properties, such as texture and degrees of overlap. The segmentation performance of ASPM demonstrates that the contour/shape based deformation procedure yields good segmentation accuracy and can be successfully adapted to numerous cytology images. Compared to the other segmentation models for overlapping cells in the literature, the proposed ASPM offers quicker with more precise segmentation for highly overlapped cells in different cervical images without extensive modifications. These results imply that the proposed ASPM approach is a promising choice for integration into a fully automated cervical cancer screening system.

Cytology image shape prediction model shape priors overlapping cell segmentation

Abstract doi.org/10.14569/IJACSA.2026.0170875 PDF

76

Black-Box Cybersecurity Assessment of Software as a Medical Device (SaMD): A Case Study

Author 1: Abdulmajeed Alshammari Author 2: Shouki A. Ebad

Based on Sommerville’s robust software reliability theory and eight design principles based on best practices (DPG), this study conducted a comparative assessment of two Dexcom software platforms: the Dexcom Clarity web portal and the Dexcom ONE+ iPhone application. The first phase of the study employed a “black-box” assessment framework, combining off-the-shelf security analysis tools with application-layer behavioral testing workflows developed specifically for this project. The initial two phases focused on the web portal, analyzing communication encryption and HTTP security header configurations. Subsequent phases involved a more in-depth implementation assessment, focusing on actual interactive features, data types, and automated responses triggered by glucose sensor alerts within the application. However, this approach also revealed that certain system elements required access to the source code for verification. The resulting workflow enables the assessment of all externally visible security attributes while identifying system components that necessitate source code access for validation. Of the eight assessment criteria, four were found to have issues, either in whole or in part. Key issues identified include: session cookies missing the SameSite attribute (G1); acceptance of clinically abnormal carbohydrate values without rejection or warning (G2); inclusion of the unsafe-inline directive in the Content Security Policy (CSP), increasing the risk of Cross-Site Scripting (XSS) attacks (G4); and the absence of a dedicated alert mechanism for Wi-Fi connectivity loss during critical medical functions (G7). Due to the requirement for source code access, this “black-box” approach could not evaluate the remaining four criteria (G3, G5, G6, and G8). Finally, this paper presents a re-producible evaluation procedure and offers applicable, standards-compliant mitigation recommendations for each identified issue.

SaMD DPG black-box testing dexcom clarity dexcom ONE+ input validation CSP XSS CSRF mHealth security

Abstract doi.org/10.14569/IJACSA.2026.0170876 PDF

77

LEAOT: Load-, Deadline-, Latency-, and Energy-Aware Task Offloading for Scalable IoT Edge–Cloud Systems

Author 1: Ayman Noor

Internet of Things (IoT) applications increasingly offload sensing and analytic tasks to edge and cloud resources. Cloud processing offers high computational capacity but can increase round-trip delay and wireless energy consumption, while edge processing reduces access delay but can suffer from queue buildup when many devices select the same nearby node. This study proposes LEAOT: Load-, Deadline-, Latency-, and Energy-Aware Task Offloading for Scalable IoT Edge–Cloud Systems, a lightweight load- and deadline-aware task offloading method for IoT edge–cloud systems. Unlike black-box learning methods, LEAOT uses an explainable online score that combines the estimated communication delay, the First-Come First-Served (FCFS) aggregate queue delay, the execution delay, the device-side energy, the soft deadline pressure, and a dimensionless infrastructure-pressure term. The method also uses an exponentially weighted link estimate and a virtual workload correction step so that stale bandwidth and queue estimates are not treated as fixed constants. A controlled discrete-event evaluation is conducted across 10 independent trials, heterogeneous task sizes, and scalable edge topologies ranging from 2 to 16 edge nodes. Results show that LEAOT reduces deadline violation to 0.69% and maintains low energy of 0.117 J/task under the reference setting, while remaining competitive with delay-resource and drift-plus-penalty baselines. The results also quantify the effect of node scalability and the resource-pressure weight.

Internet of Things edge computing cloud computing task offloading latency energy efficiency deadline-aware scheduling

Abstract doi.org/10.14569/IJACSA.2026.0170877 PDF

78

Calendar and Session Representations in Mixed-Calendar Multivariate Forecasting

Author 1: Khudran M. Alzhrani

Multivariate forecasting with variables observed on different temporal calendars requires a choice about how their histories are represented in a common input sequence. Prior work has largely treated mixed-calendar alignment as a preprocessing step or focused on modeling irregular observations, leaving direct comparisons of alternative temporal representations under identical forecast targets and evaluation dates less examined. This study compares a Calendar representation, which preserves the complete daily grid, with a Session representation, which retains only observed financial trading sessions. Epidemiological and financial data from Saudi Arabia and Germany are evaluated using identical forecast targets and target dates. Under fixed-duration matching, both representations use the same preceding 28 calendar days, whereas under fixed-sequence-length matching, Calendar uses 20 calendar days and Session uses the 20 most recent trading sessions. Ridge regression, a multilayer perceptron, and a gated recurrent unit are used to examine whether representation effects vary across model classes, while additional variants evaluate observation indicators and elapsed-time information. Ridge consistently favors Calendar across the evaluated empirical settings, MLP generally favors Calendar with variation across countries and matching rules, while GRU effects are smaller in magnitude and most 95% uncertainty intervals span zero. The choice between fixed-duration and fixed-sequence-length matching changes some representation effects, while adding observation indicators or elapsed-time information does not consistently improve performance. A target-specific naive baseline outperforms all core learned configurations, so the learned-model comparisons are interpreted primarily as analyses of representation sensitivity. In controlled simulation, representation differences are small under regular scheduled closures but become more Calendar-favoring when additional observation gaps are introduced. Overall, the results show that the effects of temporal representation vary with the model, the construction of the historical input, and the way observation gaps arise.

Mixed-calendar forecasting temporal representation multivariate time-series forecasting temporal alignment ir-regular time series observation gaps

Abstract doi.org/10.14569/IJACSA.2026.0170878 PDF

79

Code Smell Severity in Predictive Modeling for Software Quality Risk and Maintenance Effort: A Systematic Literature Review

Author 1: Aamina Banu Author 2: Thenuri Hettiarachchi Author 3: Chaman Wijesiriwardana

Code smells are structural indicators of poor soft-ware design that have been empirically associated with higher defect rates and increased maintenance effort. Although research on smell detection, software quality risk, and maintenance effort estimation has grown significantly, these areas have largely been studied separately. In addition, the role of smell severity in predictive modeling remains insufficiently synthesized. This Systematic Literature Review (SLR) investigates empirical studies published between 2010 and 2025, while also including selected foundational studies where appropriate. A total of 38 studies were selected for analysis through a structured search and screening process. The review addresses two research questions: 1) how smell type and severity influence software quality risk and maintenance effort, and 2) the extent to which current predictive approaches incorporate smell-related information in software defect prediction. The findings show that design-level smells are more consistently associated with fault-proneness than method-level smells. The review also finds that severity-aware models generally outperform binary detection models. However, the independent effect of code smells on directly measured maintenance effort remains inconsistent when confounding factors such as code size and code churn are taken into account. Furthermore, fully unified models that jointly predict defect risk and maintenance effort using smell severity as structured input remain largely absent. Overall, these findings highlight the need for standardized severity operationalization, integrated datasets, and interpretable unified modeling frameworks to advance software quality analysis.

Code smells maintenance effort software quality risk smell severity software quality

Abstract doi.org/10.14569/IJACSA.2026.0170879 PDF

80

Indie at Scale: How Large Language Models Make Community-Centered Game Development Economically Viable

Author 1: Jusuf Qarkaxhija

Large language models are now widely used in software engineering, yet most evidence of their practical effect comes from corporate productivity studies or synthetic bench-marks. This study offers a complementary view from the solo Indie developer’s perspective: a reflective case study of two iOS games developed in continuous pair programming with a large-language-model coding assistant and subsequently released on the App Store. Both games target communities that mainstream consumer software has long underserved. The Listening Maze is a spatial-audio echolocation maze for blind and low-vision players. Trace Memory Game is a memory-recall game built around brief-flash visual encoding and freehand redraw, positioned for daily cognitive engagement rather than cognitive training or clinical intervention. We describe the development methodology, the four-step human-LLM loop used throughout, the resulting codebases, an indicative break-even analysis, and what the experience implies for socially oriented software development by very small teams. We propose the crossover thesis as a design heuristic and label every claim in the discussion as observation, interpretation, or hypothesis. The evidence is an existence proof from a single developer on a single platform, not a population estimate.

Large language models AI-assisted software development mobile accessibility spatial audio cognitive engagement human-AI collaboration Indie games

Abstract doi.org/10.14569/IJACSA.2026.0170880 PDF

81

Vocabulary Without Infrastructure: A Bibliometric Analysis of Explainability and Equity in AI-Based Personalised Learning

Author 1: Ahmed Echchoayeby Author 2: Wafae Abbaoui Author 3: Reida Ilal Author 4: Nassim Kharmoum Author 5: Soumia Ziti

Artificial intelligence has become a central paradigm in personalised learning. However, it remains unclear whether the two principles that responsible AI treats as foundational, explainability and equity, are integrated into the field’s intellectual architecture or are only loosely associated with it. This question is addressed through a bibliometric analysis of 1371 publications indexed in Scopus and Web of Science between 2015 and 2025. Performance analysis and science-mapping techniques were applied using bibliometrix and VOSviewer, with cluster structure quantified through modularity and inter-cluster density metrics, and uncertainty assessed through 1000-iteration edge-resampling bootstrap. Keyword co-occurrence analysis showed that ethics-equity vocabulary connects to the applied-AI cluster at densities comparable to those of technical vocabulary (0.20 vs 0.21), but the ethics-equity and technical streams interact directly only at half that rate (0.11; bootstrap 95% CIs non-overlapping), engaging primarily through the applied-AI mediator cluster rather than with each other. Co-citation analysis showed that the foundational responsible-AI references in education do not register as cluster-forming nodes. No specialist responsible-AI venue appears in the field’s Bradford core, and Sub-Saharan Africa and Latin America remain almost absent from the country-level collaboration network. To account for this pattern of vocabulary diffusion without infrastructural consolidation, we propose the Structural Integration Model (SIM). The SIM is a three-proposition framework that specifies the conditions under which explainability and equity can become structurally embedded in the field’s foundation rather than only thematically present within it.

Bibliometric analysis AI in education personalised learning explainable AI educational equity responsible AI science mapping

Abstract doi.org/10.14569/IJACSA.2026.0170881 PDF

82

Improving Flight Delay Prediction Using Integrated Departure Disruption, Airline-Route Stability, and Schedule Intensity Features

Author 1: Huthaifa Aljawazneh

In the aviation industry, flight delays represent a major challenge due to their economic impact, operational disruptions, and adverse effects on passenger satisfaction and transportation efficiency. This study proposes an integrated feature engineering framework for flight delay prediction that combines three groups of engineered features, namely Departure Disruption Features (DDFs), Airline-Route Stability Features (ARSFs), and Schedule Intensity Features (SIFs), with the original flight delay dataset. These feature groups are designed to capture complementary operational, historical, and scheduling characteristics that are not directly represented in the original data. Five machine learning classifiers were evaluated across five experimental scenarios. These included the original feature set, the original feature set combined separately with each engineered feature group (DDFs, ARSFs, and SIFs), and the original feature set combined with all three engineered feature groups. This design enabled assessment of the individual and complementary contributions of the proposed features to flight delay prediction. The results indicated that the engineered features improved prediction performance across the evaluated configurations. In particular, combining all three engineered feature groups with the original feature set outperformed the baseline feature set and each individually combined feature-set configuration, supporting the complementary predictive value of the proposed engineered features. The complete feature set with LightGBM and SMOTE-Tomek achieved the best performance, with an accuracy of 0.9820 and an AUC of 0.9884, compared with 0.9487 and 0.9663, respectively, for the baseline feature set.

Flight delay prediction feature engineering aviation analytics class imbalance machine learning

Abstract doi.org/10.14569/IJACSA.2026.0170882 PDF

83

Deep Autoencoder and Deep Q-Network-Based Feature Learning with Multi-Head Residual Attention DNN for Cyberattack Classification

Author 1: Dharani Kanta Roy Author 2: Hemanta Kumar Kalita

As society’s digital infrastructure grows, so do the threats. We cannot deny internet access, so we need to safeguard ourselves against cyber threats. Every day, new threats emerge. The most conventional systems classify cyberattacks in a closed set; that is, training and testing use the same set of attack behaviors. In the real world, we cannot train our system on all attack patterns because attack behavior evolves daily. To address this issue, this study proposes open-set cyberattack classification, in which we train our system on a limited set of labeled attack types and, during testing, use a larger unlabeled data sample that includes both known and unknown attack patterns. Our framework integrates a Deep Autoencoder (DAE)-based latent representation, an Explainable Deep Q-Network (Explainable DQN) for feature selection, and a Multi-Head Residual Attention Deep Neural Network (MHRA-DNN) for final classification. The DAE transforms high-dimensional network traffic into a low-dimensional latent feature representation. This reduces redundancy while preserving important feature characteristics. The Explainable DQN selects the features that contribute most to the model’s performance. It also improves explainability by highlighting the most influential features. The selected features are then processed by the MHRA-DNN. It captures multiple feature combinations and strengthens learning through attention mechanisms and residual connections. We evaluate the proposed framework on the benchmark cybersecurity datasets UNSW-NB15, APA-DDoS, and CICIDS2017. We set up an open-set environment that involves known and unknown attack classes. The proposed model performs strongly across accuracy, precision, recall, F1-score, specificity, MCC, and NPV. In addition, parameter tuning improves model stability, and SHAP-based analysis provides evidence of feature contributions to model performance. Overall, the proposed model offers an adaptive, explainable, and robust solution to cyberattack classification in an open-set environment.

Cyberattack classification deep autoencoder feature selection deep q-network multi-head attention deep neural network SHapley additive exPlanations

Abstract doi.org/10.14569/IJACSA.2026.0170883 PDF

84

Hybrid Approach Combining Markov Chains, HMM, and Bidirectional GRU to Detect Blockages in Programming

Author 1: Grota Abdelkader Author 2: Mohammed Erritali Author 3: Patrick Etcheverry Author 4: Thierry Nodenot

A student stuck in a failing compile loop looks productive from across the room. The instructor sees keystrokes; the blockage is invisible until the student raises a hand or falls silent. Existing detection methods work at session level: they report how many compilations occurred, not what happened between them. Sub-minute resolution is needed. We built FusionAdaptative, a hybrid model that reads raw keystroke traces at 30-second windows, combining Markov Chains on a 13-state behavioral taxonomy, HMM latent state inference, and BiGRU with attention, combined at the level of representation. On 70 first-year CS students, 287,236 actions, 220 sequences, a counter-intuitive result holds: all eight evaluated configurations, from a non-neural Random Forest to the full hybrid, reach an equivalent Macro-F1 of about 90.7% (Friedman χ2(7) = 3.53, p = 0.83; Nemenyi post-hoc: no pairwise difference, critical difference = 4.70). This is not a weakness of the model but a property of the design: once cognitive states are defined by quantitative thresholds validated at κ = 0.81, classification is constrained enough that architecture no longer matters. The discriminative power is in the taxonomy, not the complexity of the model. FusionAdaptative’s contribution is temporal. It detects blockages on average 2.8 minutes before they become visible, a lead that only sequential modeling (BiGRU attention and HMM Viterbi decoding) can produce and that aggregated classifiers such as Random Forest cannot. The lead varies by blockage type: 3.2 minutes for conceptual, 2.6 for syntactic, 2.4 for strategic, each anticipated earliest by a different component of the model.

Attention mechanism bidirectional GRU blockage detection early detection educational data mining

Abstract doi.org/10.14569/IJACSA.2026.0170884 PDF

85

A Database-Driven Architecture for Integrating and Serving 3D Condominium and Cadastral Information: The CondoMaps Platform

Author 1: Tarawut Boonlua Author 2: Kritsanu Palopakorn Author 3: Sontaya Ratanatip Author 4: Somporn Wongjampa Author 5: Praewpan Pasanam Author 6: Viparat Nupattaya Author 7: Ketut Tomy Suhari

Three-dimensional condominium information systems require more than visual building models: spatial objects must be constructed from authoritative plans and linked consistently to cadastral, registration, building, unit, parcel, and juristic-person records. This study presents CondoMaps, a database-driven architecture for constructing, organizing, and serving 3D condominium information in Thailand. The workflow transforms verified floor plans and cadastral references into georeferenced LOD2 building and unit geometries, assigns persistent project–building–floor–unit identifiers, stores administrative and legal attributes in a normalized relational database, and publishes the integrated objects through controlled map, scene, query, and REST services. The accepted production snapshot contains 7,150 condominium projects, more than 180,000 units, over 650 LOD2 buildings, at least 6,500 contextual LOD1 buildings, 194,304 spatial features, and 457,936 linked attribute records. The combined database and file repository occupies approximately 6 TB and supports 48 published services across 22 land-office branches. Quality checks on the accepted snapshot found complete mandatory documentation and fields, no duplicate normalized keys, and no orphan records in the reported integrity tests. Linkage rates of 100% apply only to the specified sets of 7,150 evaluated records or objects and are not estimates for all units. Horizontal RMSE values of 0.0775 m and 0.031674 m were obtained in two four-control-point plan rectification cases; they are case-level results, not evidence of repository-wide positional accuracy. A browser implementation demonstrates retrieval of associated unit, juristic-person, coordinate, title-deed, and parcel-survey information from a selected 3D building. The contribution is the traceable end-to-end linkage among source documents, normalized records, identifiable 3D objects, and governed web services rather than visualization alone. Performance and formal security benchmarks remain future work.

3D Condominium model spatial database 3D cadastre data integration web GIS service-oriented architecture CondoMaps

Abstract doi.org/10.14569/IJACSA.2026.0170885 PDF

86

Uncertainty Adaptive Shared Control with Multimodal Intent Fusion for Human-Guided Quadruped Locomotion on Uneven Terrain

Author 1: Likai Wu

Human-guided quadruped locomotion requires re-liable intent inference and safe authority allocation under un-certain interaction and uneven terrain. This study proposes an uncertainty-adaptive shared-control framework with multimodal intent fusion for a Unitree Go2 quadruped. Interaction force, relative human–robot motion, operator-motion cues, and terrain context are fused through a temporal model to estimate desired motion and maneuver mode. Intent confidence is evaluated using posterior entropy and prediction variance, while locomotion risk is computed from traversability, slope, obstacle proximity, interaction load, and stability margin. The human authority factor is then adjusted online to blend the inferred human command with a robot-safe command. The shared command is realized through gait scheduling, terrain-aware foothold optimization, whole-body control, and interaction-compliant base regulation. Experiments across five guidance scenarios show that the proposed method improves intent accuracy, interaction load, slip resistance, stability margin, and task success compared with force-only guidance, fixed-authority shared control, and ablated variants.

Quadruped robot human–robot interaction shared autonomy intent recognition uncertainty estimation terrain-aware locomotion assistive robotics

Abstract doi.org/10.14569/IJACSA.2026.0170886 PDF

87

Predicting PM10 from PM2.5 and NO2 Levels in Metropolitan Lima, Peru: A Regularized Regression Approach

Author 1: Jorge Chavarri Centeno Author 2: Hilter Lanche Cerron Author 3: Joel Rivera Matias Author 4: Juan Jesús Soria-Quijaite Author 5: Nemias Saboya

Suspended particle pollution (PM10) represents a serious public health problem in Metropolitan Lima, since its high levels are related to an increase in respiratory and heart diseases. This work aims to anticipate PM10 concentrations in seven districts of Lima using a multiple linear regression (OLS) model with the Ridge and Lasso regularization techniques. PM2.5 and NO2 were selected as predictors because they were the only pollutant variables available in the SENAMHI monitoring dataset; hourly records collected between 2015 and 2024 were used, yielding 169,308 observations after removing missing values and outliers. The data were separated into 80% for training and 20% for validation, applying cross-validation to determine the most appropriate regularization parameter. The final model reached a coefficient of determination (R2) of 0.23 and a mean squared error (MSE) of 521 in the test set. The findings indicate that PM2.5 is the main factor that predicts PM10, while NO2 exerts a secondary impact. Although the predictive capacity of the models is limited, the use of both Ridge and Lasso helped stabilize the coefficients and minimize overfitting. Given this limited explanatory power, the model should be regarded as an initial, interpretable baseline for PM10 monitoring in Metropolitan Lima rather than an operational forecasting tool. It is suggested to include meteorological variables and investigate non-linear models in future studies to enhance the accuracy and usefulness of the model in air quality management.

Multiple linear regression particulate matter of 10 micrometers or less (PM10) particulate matter of 2.5 micrometers or less (PM2.5) nitrogen dioxide (NO2) ridge regularization Lasso regularization

Abstract doi.org/10.14569/IJACSA.2026.0170887 PDF

88

Conformal, Explainable Machine Learning for Forensic Malware Triage and IoT Attribution

Author 1: Arafat Al-Dhaqm Author 2: Abdullah Alajmi

Machine learning can triage digital evidence at scale, but two obstacles limit its forensic adoption: opaque decisions, and point predictions without a valid statement of confidence. We present ForensiQ, a hierarchical and explain-able framework that pairs a two-stage detect-then-attribute pipeline with split-conformal prediction. Conformal calibration is distribution-free and turns classifier scores into prediction sets with finite-sample coverage guarantees, so an investigator can bound the attribution error rate before it enters a report. We evaluate on two public corpora: CIC-MalMem-2022 memory forensics (58,596 samples) and TON_IoT telemetry (401,119 records, seven sensors). All results are the mean of five seeded runs on commodity hardware. ForensiQ reaches 99.99% binary triage accuracy and 87.4% four-class family attribution (macro F1 0.811), matching the state of the art. Empirical conformal coverage is 90.2%, 95.1%, and 99.0% against 90%, 95%, and 99% targets, with prediction sets of 1.08 to 1.58 labels. A class-conditional analysis shows the guarantee is marginal rather than per-family; per-family Mondrian calibration restores 95%validity at a modest set-size cost. On TON_IoT, binary detection averages 85.8% and reaches 100% on the garage-door sensor. Telemetry-only attack typing is weakly identifiable on several devices, a negative result we quantify, and it depends partly on timestamp features that we ablate and report as a primary condition. ForensiQ offers a reproducible, uncertainty-aware baseline for AI-assisted forensics, and we release the full configuration needed to regenerate every table and figure.

Digital forensics memory forensics malware attribution conformal prediction explainable artificial intelligence IoT security machine learning

Abstract doi.org/10.14569/IJACSA.2026.0170888 PDF

89

From IoT Vulnerabilities to Intrusion Detection: An Explainable Vulnerability-Aware Machine Learning Framework for Smart Home IoT Security

Author 1: Huda Aldawghan Author 2: Mounir Frikha

The swift deployment of IoT-based smart home appliances has increased the attack surface for the smart environment and exposed it to attacks like botnet command-and-control communications, brute force attacks, denial-of-service attacks, and web-based attacks. Even though the Intrusion Detection Systems (IDSs) that use Machine Learning (ML) algorithms achieve a very high detection rate, most existing solutions focus on predictive performance but lack the ability to link detected attacks to the corresponding vulnerabilities in the Internet of Things (IoT). In this paper, an interpretable vulnerability-aware ML-based approach is presented to solve this problem through the integration of vulnerability classes of IoT, attack classes, network flow attributes, and ML features into one interpretation model. The proposed method uses leakage-aware pre-processing, addressing class imbalance, and compares Random Forest, XGBoost, and soft voting ensemble ML techniques using the CSE-CIC-IDS2018 dataset. Experimental outcomes indicate that XGBoost outperforms the other approaches in terms of performance, with an accuracy of 98.22%, precision of 99.69%, F1-score of 95.36%, ROC-AUC of 99.07%, and only 760 false alarms, which is approximately 19× lower number of false positives compared to Random Forest while keeping a similar level of detection efficiency. In addition to numeric assessment of the approach performance, the suggested model provides the vulnerability-oriented interpretation module that establishes mapping between prominent network flow attributes and possible IoT vulnerability states and attacks. Therefore, the integration of an interpretable vulnerability reasoning component into a high-performing tree-based machine learning algorithm proves to be effective for smart home IoT intrusion detection.

IoT security intrusion detection system machine learning XGBoost random forest vulnerability-aware security botnet detection smart home explainable AI network-flow analysis

Abstract doi.org/10.14569/IJACSA.2026.0170889 PDF

90

A Trust-Aware Federated Learning Framework Based on Zero Trust Architecture for Secure Cloud Intrusion Detection

Author 1: Eman M. Mohamed Author 2: Suzan S. Basloom

Cloud computing provides scalable infrastructure but introduces critical challenges to data privacy, trust management, and intrusion detection. To address these issues, we present a trust-aware framework that integrates a Federated Learning (FL)-based distributed ensemble with Zero Trust Architecture (ZTA) for secure cloud environments. The framework enables collaborative learning while keeping data local and private. Zero Trust is implemented by periodically verifying identity and evaluating client contributions with a lightweight local trust filter before aggregation. For reliability and agreement with the local model, an ensemble of Random Forest and Decision Tree classifiers is used. Contributions are only aggregated when satisfying the predefined trust and performance thresholds, thereby reducing the influence of potentially unreliable or malicious inputs under the assumed threat model. The proposed framework is evaluated on the UNSW-NB15 dataset under non-IID client data distribution, achieving 96% accuracy, 95.1% precision, 94%recall, 0.96 AUC-ROC, and an average per-round latency of 180 ms. Comparative results demonstrate that the framework provides a better trade-off between detection accuracy, privacy preservation, and trust enforcement than several state-of-the-art baselines, positioning it as a practical solution for cloud intrusion detection.

Cloud security federated learning zero trust architecture intrusion detection system trust filtering mechanism decentralized learning

Abstract doi.org/10.14569/IJACSA.2026.0170890 PDF

91

A Heterogeneity-Aware Federated Learning Framework for Graph Neural Networks in Multi-Hospital Medical Systems

Author 1: Chethana Prasad Kabgere Author 2: Shylaja S S

Medical institutions increasingly hold data that is relational rather than tabular: patients linked by medical history, diagnostics by dependency, and doctors by consultations. Graph neural networks (GNNs) are a natural fit for such data, but hospitals cannot pool it directly, since records remain confined by law to the collecting hospital. Federated learning lets a central server train a shared model by aggregating parameter updates instead of raw graphs. GNN parameters, however, define a graph-dependent message-passing operator, so when hospitals differ in topology, degree distribution, and homophily, standard update-averaging no longer produces a coherent update for any hospital’s actual operator, a mismatch accuracy metrics can mask, hitting structurally atypical hospitals hardest. We introduce the Global Geometric Reference Structure (GGRS), a server-side layer that regulates updates before aggregation: directional soft-weighting down-weights updates that disagree with the hospital consensus, subspace projection keeps only shared directions, and sensitivity clipping stops any one hospital from dominating the aggregate. Across six benchmark graphs, two architectures, and four federated optimizers, GGRS improves accuracy by up to 5.1% for structurally under-represented hospitals, with the best-performing combination, SCAFFOLD paired with GGRS, also reaching target accuracy up to 12 rounds faster than unregulated aggregation. These results suggest geometry-aware aggregation helps federated GNNs serve heterogeneous hospitals more equitably, without requiring any hospital to change how it trains locally.

Inter-hospital federated learning medical graphs geometric coherence message-passing operators structural het-erogeneity

Abstract doi.org/10.14569/IJACSA.2026.0170891 PDF

92

DISR: Distributional Item Representations for Sequential Recommendation

Author 1: Abdelilah BAJJOU Author 2: El Habib NFAOUI

Sequential recommender systems predict the next item from a chronological interaction history, typically representing users and items as point embeddings scored by dot product. DISR (Distributional Item Representations for Sequential Recommendation) instead represents items and sequence queries as diagonal Gaussian distributions. Item means are initialized from frozen Sentence-BERT text representations, while query and item variances are learned during training. Candidates are scored using closed-form symmetric KL divergence within the standard sampled-softmax cross-entropy framework. Experiments on Amazon Reviews 2023 Beauty and Sports evaluate 50K-user and full-scale settings, with five seeds at 50K. DISR improves NDCG@10 over the strongest non-DISR baseline by +12.8% on Beauty 50K and +15.6% on Sports 50K. At full scale, DISR leads on Sports by +2.6% over DuoRec and remains within 0.8% of the best NDCG@10 on Beauty. The largest gains occur for users with one to three interactions, reaching +14.5% on Beauty and+19.3% on Sports. DISR also converges substantially faster than the strongest contrastive baselines at full scale. Ablation results support the contribution of the overall distributional formulation, while the full-scale evaluation remains single-seed.

Sequential recommendation distributional embedding Kullback–Leibler divergence cold-start recommendation text-initialized embeddings amazon reviews 2023

Abstract doi.org/10.14569/IJACSA.2026.0170892 PDF

93

Boosting Versus Foundation Models for Ordinal Home-Sharing Engagement Classification in Latin America

Author 1: Pedro Shiguihara Author 2: Nils Murrugarra

Review activity is the standard public proxy of guest engagement on peer-to-peer accommodation platforms, yet is modeled almost exclusively as count regression. This study presents a three-level engagement classification of peer-to-peer home-sharing in Latin America: 76,624 listings from Buenos Aires, Santiago, and Mexico City, with tercile labels from training folds only. A pre-specified, amendment-logged evidence ladder of fifteen attempts plus three contingency campaigns measures which interventions affect macro-averaged F1. A tuned LightGBM over engineered features reaches average macro F1 0.8249, 23.0 points above the best cold-start result (0.5945); a feature-level temporal audit lowers the deployable cold-start reference to 0.5301, widening the gap to 29.5 points. Three findings emerge. First, every full-scenario result lands in a narrow band (0.81–0.83) that doubled budgets do not move: an observed performance plateau within the evaluated feature and model space. Second, Spanish text helps in the cold-start scenario (+2.4 points nominally, +5.3 over the audited strict set) but adds little once review-derived features are present, and sparse lexical vectors match dense multilingual E5 embeddings at far lower cost. Third, an untuned tabular foundation model reaches 0.8205 under a pre-specified 10,000-row context subsample. A clearly labeled post-hoc re-evaluation at full context, outside the pre-specified protocol, reaches 0.8266, the best average observed, which an exploratory multi-resampling confirmation upholds in all twelve paired comparisons. Ordinal metrics, per-class breakdowns, and a Frank–Hall baseline locate the residual error in the middle tercile; repeated outer resampling bounds selection optimism at 0.2–0.3 points.

Peer-to-peer home-sharing ordinal classification engagement prediction gradient boosting multilingual text embeddings hyperparameter optimization tabular foundation models TabPFN Latin America

Abstract doi.org/10.14569/IJACSA.2026.0170893 PDF

94

A User-Intent Context-Aware Recommendation Approach for Low-Resource Online Learning Environments (Mauritanian Context)

Author 1: Massra Sabeima

User experience in distance learning is often conceptualized through interactions between the learner and the system’s interface. These interactions can be understood as the dynamic interplay between user behavior and system behavior, each initiated or guided by distinct mechanisms. User behavior, inherently autonomous, cannot be directly dictated, whereas system behavior can be designed to guide or constrain user actions. In human-guided systems, the process primarily responds to user input, offering limited control over the learning experience. Conversely, system-guided interactions selectively present content and collect targeted information, shaping the learner’s engagement and, in some cases, eliciting responses that the user may not consciously provide. This duality of guidance highlights the critical role of system design in influencing both overt and latent aspects of learner behavior, emphasizing the importance of adaptive mechanisms to optimize user experience in distance learning environments.

User intent modeling context-aware recommendation low-resource online environments

Abstract doi.org/10.14569/IJACSA.2026.0170894 PDF

95

Beyond Model-Centric AI: Embedding Governance as an Architectural Component with Measurable Efficiency

Author 1: Waleed Al Shehri

Production AI systems typically apply governance—compliance checking, auditing, and explanation—as an external control layer that observes outputs after they are produced. In this paper, embedded governance is proposed: a governance gate that forms part of the system architecture itself, positioned between inference and release, so that no output leaves the system without a measurable governance assessment. For each candidate output, a governance score is computed from three factors that are checked mechanically rather than judged by a model: intent conformance, formulation traceability, and verification. On the basis of this score, each output is released, remediated, or blocked. Governance efficiency η, the ratio of intercepted error cost to governance cost, is further defined, by which the claim that governance pays for itself becomes falsifiable rather than asserted. Weights and thresholds are calibrated by maximizing η on a validation split and reported on held-out data. Evaluation uses the full credit-card fraud detection dataset of 284,807 transactions under a temporal split, so that the evaluation period follows the training period and distribution drift arises naturally. The calibrated gate intercepts 99% (95% CI 94.6–99.8) of severely shifted requests at a mean overhead of 3.74 ms, and η reaches 4.52 when releasing unreliable outputs carries cost, while remaining below break-even under accuracy-only cost models. A severity sweep locates the detection boundary and shows interception falling to 2% under mild shift. An ablation identifies distributional checks as the load-bearing components and reveals that η can be inflated by a gate that governs less, so efficiency must be read alongside coverage. Under identical workload, external monitoring and post-hoc assessment detect the same errors but prevent none, whereas inline placement reduces released error cost by 82.

AI governance software architecture embedded governance governance efficiency ablation study failure semantics out-of-distribution detection MLOps

Abstract doi.org/10.14569/IJACSA.2026.0170895 PDF

96

Proportional Reward and Temporal Discounting for Monopoly-Free Heterogeneous Metaheuristic Portfolios

Author 1: Bilal Bataineh Author 2: Sofian Kassaymeh

Real-world optimization landscapes are typically dynamic, high-dimensional, and uncertain, and a single meta-heuristic with fixed control parameters rarely sustains strong performance across such environments, as formalized by the No Free Lunch theorem. Existing adaptive frameworks attempt to address this through online operator or algorithm selection, but they suffer from two persistent limitations: coarse feed-back mechanisms that reward the frequency rather than the magnitude of improvements, and cumulative memory bias that allows early-performing algorithms to monopolize selection long after their advantage has faded. This work proposes a problem-agnostic adaptive framework that integrates the Relative Improvement Metric (RIM), a proportional reward quantifying the magnitude of each solver’s contribution, with a Sliding Window (SW) policy that discounts older rewards temporally so that algorithmic influence remains contingent on recent effectiveness. The framework orchestrates a heterogeneous portfolio of eight metaheuristics (GA, PSO, GWO, ACO, SSA, ABC, WOA, FA) through probabilistic selection driven by SW-RIM weights, with a small base probability that prevents any solver from being permanently excluded. Empirical evaluation across 23 standard benchmarks and the CEC2020 suite shows that the proposed mechanism reduces maximum solver participation from above 70% in the baseline configuration to under 40%, improves mean fitness on the majority of functions in both groups, and achieves statistically significant gains over both a baseline portfolio and a sliding-window-only variant (Wilcoxon p < 0.001; Vargha-Delaney A12 between 0.72 and 0.85, large effect across all comparisons). The two exceptions are functions with deceptive or ill-conditioned landscapes (a narrow-valley Rosenbrock-type function and a highly multimodal Schwefel-type function), where all three configurations perform comparably, indicating that SW-RIM’s benefit is contingent on the portfolio containing at least one solver structurally suited to the current landscape rather than on the selection strategy alone. The results support SW-RIM as a lightweight, general-purpose mechanism for sustaining diversity and impact-sensitive adaptation in complex continuous optimization, without the training cost of reinforcement-learning-based selectors.

Adaptive metaheuristic relative improvement metric sliding window hybrid optimization algorithmic selection dynamic adaptation

Abstract doi.org/10.14569/IJACSA.2026.0170896 PDF

97

Focal-Loss Transformer with Feedback-Validation Learning for Imbalanced Drug Interaction Extraction

Author 1: Hiba Chanaa Author 2: El Habib Nfaoui Author 3: Chafik Boulealam Author 4: Chakir Loqman

Drug-drug interactions (DDIs) are a leading cause of preventable adverse drug events, and the standard benchmark corpus for extracting them from biomedical text is severely im-balanced towards non-interacting pairs, making rare interaction types hard to learn. We asked whether combining focal loss, entity-aware transformer encoding, and feedback-guided feature fusion in a single architecture could improve minority-class DDI extraction under this imbalance while requiring less training data than prior methods. We propose FoLT-DMCNN-FBVL, which integrates a BiomedBERT backbone with entity marker injection, a multi-branch neural classifier, and feedback-based validation learning, trained on a small balanced subset of the SemEval- 2013 DDIExtraction corpus and evaluated on the full blind test set. FoLT-DMCNN-FBVL matched the most recent state-of-the-art result (statistically indistinguishable; nominal absolute+0.33%) and significantly outperformed the widely cited CNN-DDI baseline (absolute +4.32%), while using less than one-fifth of the available training data. These findings show that a doubly imbalance-aware transformer architecture can match or exceed current state-of-the-art DDI extraction performance with substantially greater data efficiency. Such architectures could support more scalable and cost-effective pharmacovigilance and clinical NLP pipelines in settings where annotated data for rare interaction types is limited.

Drug-drug interaction extraction biomedical natural language processing transformer focal loss class imbalance pharmacovigilance

Abstract doi.org/10.14569/IJACSA.2026.0170897 PDF

98

Deep Learning-Based System for Accurate Sleep Stage Classification and Early Detection of Sleep Disorders Using EEG and EOG Signals

Author 1: Ramesh G Author 2: Virgil Popescu Author 3: Cristina Sultanoiu (Patularu) Author 4: Vaikunta Pai T Author 5: Gabriela Ana Maria Lupu (Filip) Author 6: Md. Abul Kalam Azad Author 7: Ramona Birau Author 8: Roxana Mihaela Nioata (Chireac)

Sleep is essential for maintaining overall physical and mental health, yet analyzing sleep patterns manually is a com-plex and time-consuming process that requires expert knowledge and is often prone to subjectivity. In this study, an automated and efficient deep learning-based approach is developed to classify sleep stages using the Sleep-EDF dataset. Physiological signals such as Electroencephalogram (EEG), Electrooculogram (EOG), and Electromyogram (EMG) are pre-processed and segmented into fixed-length epochs for analysis. These segments are then fed into models including Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and a hybrid CNN-LSTM architecture, which effectively capture both spatial and temporal features of the signals. The system classifies sleep into stages such as Wake, N1, N2, N3, and REM accurately and robustly. Among the implemented models, the proposed hybrid CNN-LSTM model achieved the best performance with an accuracy of 94.67%. A basic analysis of sleep stage patterns is also performed to observe possible irregularities. Overall, this work presents a scalable solution for sleep stage classification using deep learning techniques.

Sleep stage classification EEG EOG deep learning CNN LSTM sleep disorders polysomnography

Abstract doi.org/10.14569/IJACSA.2026.0170898 PDF

99

The Agentic Enterprise Capability Framework (AECF): A Governance-First Architecture for Scalable AI Agent Deployments

Author 1: KHAMMAL Adil Author 2: HAMZANE Ibrahim Author 3: MARZAK Abdelaziz Author 4: Hajar Taji Author 5: Elmostafa Erraitab Author 6: Mohamed EL WAFIQ

Enterprise AI adoption has reached a structural inflection point: while a majority of organizations have deployed generative AI, few have established mature governance models for autonomous agents. This disparity reflects a fundamental architectural gap: Multi-Agent Systems, Enterprise Architecture, AI agent deployment, and software architecture have approached agent coordination from separate disciplinary perspectives, with no single framework integrating persistent memory, semantic interoperability, orchestration, human oversight, and normative enforcement. Following Design Science Research, this article develops the Agentic Enterprise Capability Framework (AECF), a five-layer architecture structured around Context Persistence (CPL), Semantic Interoperability (SIL), Hybrid Orchestration (HOL), Human Governance Interface (HGI), and Governance Envelope (GEL). The framework introduces the co-evolution constraint: technical capability layers cannot mature independently of governance capacity. This constraint is operationalized through Context-Enriched Pre-Execution Validation (CEPEV), which grounds compliance checks in operational memory. Five architectural propositions formalize inter-layer dependencies (P1), scalability boundaries (P2), governance effectiveness (P3), performance accumulation (P4), and a governance scaling law (P5). The study contributes an integrated architectural model, propositional formalization, and validation agenda for governed enterprise AI agent deployments.

Enterprise AI agents agentic architecture design science research AI governance multi-agent systems policy-as-code

Abstract doi.org/10.14569/IJACSA.2026.0170899 PDF

100

Evaluating AI-Based Veterinary Symptom-Assessment Chatbots: A Review-Informed Framework for Safety, Triage, Usability, and Owner-Style Symptom Reporting

Author 1: Manisha Chawla Author 2: Abeer Alsadoon Author 3: Ahmed Fadhil Author 4: Mohammed Ameen Author 5: Arafat Abdulgader Mohammed Elhag Author 6: Areej Althubaity Author 7: Ahmed Hamza Osman

In the medical field, AI has made its mark, and almost every medical profession has experienced the development of a chatbot to check symptoms and provide instructions to the owners at an early stage. However, determining the safety and reliability of these tools will be challenging since pets are unable to self-report their symptoms, and the input for the chatbot will often be from the pet owner's perception. This may make it difficult to interpret symptoms, assess urgency, and communicate safety advice. This research proposes and adopts a systematic framework for a controlled pet health evaluation scenario to assess an AI-based veterinary symptom assessment Chatbot. The framework evaluates the chatbot's reactions in four areas – diagnostic accuracy, triage behaviour, safety communication, and usability. It is compared with the benchmark results set by experts for controlled owner-style scenarios, and a detailed analysis of the failures of this chatbot’s response is conducted using an interaction-stage error taxonomy. The framework was applied to one veterinary chatbot using 13 controlled owner-style scenarios. Each scenario was evaluated once as a single-turn interaction. The findings showed variation across the four evaluation dimensions, particularly in triage behaviour and safety communication. These findings should be interpreted as evidence from this controlled test setting and should not be generalised to all veterinary chatbots or to real-world clinical use. The study demonstrates how owner-style symptom input, expert benchmarking, multidimensional scoring, and interaction-stage error analysis can be combined within a veterinary-focused evaluation framework.

Veterinary chatbot artificial intelligence symptom assessment pet health scenarios triage behaviour safety communication usability expert benchmark

Abstract doi.org/10.14569/IJACSA.2026.01708100 PDF

101

A Real-Time IoT-Based Intelligent Health Monitoring Framework with Multi-Sensor Fusion and Automated Anomaly Detection for Continuous Cardiac Assessment

Author 1: J R Deepak Author 2: C Veerabathiran Author 3: Jamil Abedalrahim Jamil Alsayaydeh Author 4: S Padmanabhan Author 5: N Adhithan Author 6: Rex Bacarra

Continuous and real-time health monitoring is essential for the early detection of cardiovascular and physiological abnormalities in remote and resource-limited settings. This research aims to develop an intelligent IoT-based framework for multi-sensor health data acquisition, real-time processing, and automated anomaly detection for continuous cardiac and physiological assessment. The proposed system integrates multiple physiological sensors, including heart rate, SpO₂, ECG, blood pressure, and body temperature modules, interfaced with a microcontroller-based edge computing unit for real-time signal processing and decision-making. A rule-based intelligent alert mechanism is implemented to enable automated detection of abnormal conditions and trigger immediate emergency notifications via wireless communication channels including Bluetooth and GSM. Data transmission to cloud-based platforms supports remote monitoring and longitudinal health tracking. The system architecture follows a layered IoT design encompassing sensing, processing, communication, and application layers. Performance was validated through MATLAB-based simulation and experimental testing under multiple activity conditions, achieving accuracies of 92.85% for heart rate, 98.86% for SpO₂, 98.25% for systolic blood pressure, 97.37% for diastolic blood pressure, and 99.7% for body temperature. The results demonstrate that the proposed framework provides a reliable, cost-effective, and scalable solution for continuous remote health monitoring, with significant implications for smart healthcare systems and clinical decision support in IoT environments.

Internet of Things (IoT) wearable health monitoring device biomedical sensors cardiac monitoring vital sign measurements

Abstract doi.org/10.14569/IJACSA.2026.01708101 PDF

102

Dimensionality-Performance Trade-Off Analysis of the Hybrid Character-Word Embedding (HCWE) Architecture for Medical Text Classification Under Noise Conditions

Author 1: Mustazzihim Suhaidi Author 2: Rabiah Binti Abdul Kadir Author 3: Sabrina Tiun

Medical text classification is critical for applications such as automated triage and disease surveillance. However, real-world data is often corrupted by character-level noise, which substantially degrades the performance of traditional models. Although the Hybrid Character-Word Embedding (HCWE) architecture has demonstrated promising results, the optimal embedding dimensionality for balancing classification performance and computational efficiency under noisy conditions remains unclear. This study presents a comprehensive analysis of the relationship between embedding dimensionality and classification performance by evaluating the HCWE architectural family across embedding dimensions ranging from fifty to five hundred under noise levels ranging from zero percent to one hundred percent. A medical tweet dataset containing 2497 samples across 15 health categories was evaluated using a 5-fold stratified cross-validation protocol to ensure statistical robustness. The results show that HCWE-200D achieves a peak Macro F1-score of 0.860 ± 0.012 at a severe 60% noise level, exceeding the strongest baseline by 34.4% in relative performance improvement (p < 0.01, paired t-test). Direct feature-space analysis reveals that increasing dimensionality from 50-D to 500-D reduces the hash collision rate from 99.19% down to 91.90% while increasing feature vector sparsity from 2.67% to 66.33%, explaining why performance reaches a clear saturation plateau beyond 300 dimensions. Under realistic structured noise (keyboard typos, phonetic misspellings, character repetitions, and medical slang), HCWE maintains robust performance (Macro F1 = 0.248 to 0.273). Furthermore, a multi-objective deployment utility function U(D) demonstrates that HCWE-50D optimizes resource-constrained edge deployments (RAM < 15 MB, latency 0.12 ms), whereas HCWE-200D and HCWE-300D represent the optimal configurations for balanced and high-accuracy server-side deployments, respectively.

Dimensionality analysis medical text classification noise robustness hybrid embedding HCWE feature hashing statistical significance multi-objective optimization

Abstract doi.org/10.14569/IJACSA.2026.01708102 PDF

103

Adaptive Energy-Security Framework for IoT Devices Using Hybrid SPECK and Kyber Encryption

Author 1: Vegesna Siva Rama Krishnam Raju Author 2: Venkateswararao Pulipati

With the fast growth of the Internet of Things (IoT), the problem of secure and energy-efficient data exchange in resource-constrained settings has become even more challenging. The traditional cryptographic systems (RSA and ECC) are resource-consuming and susceptible to quantum attacks, and lightweight cyphers are not only less powerful but also energy-efficient. This study is a proposal of an adaptive hybrid encryption system (adjusting in real time) that dynamically utilises either a lightweight SPECK cypher or a post-quantum secure Kyber algorithm based on the energy consumption of the devices and the sensitivity of the data. The system will feature a PQC offload controller that manages key encapsulation and minimises computation latency. Simulation results in an OMNeT++ environment have shown that the proposed model is up to 45 per cent more energy-efficient and up to 40 per cent longer-lived than PQC models alone, while supporting 99 per cent security resilience against both classical and quantum attacks. Throughput levels of 1,600 packets per second are achieved by the approach, which is a trade-off between energy usage and post-quantum security, and is acceptable. It is an effective method of minimising energy use and reducing the risk of quantum attacks in IoT networks used in healthcare, industrial, and innovative city applications. With the assistance of this adaptable structure, it emphasises its relevance to energy-constrained devices.

Post-Quantum cryptography lightweight cryptography hybrid cryptography IoT security quantum-resistant networks energy-efficient encryption Kyber SPECK network lifetime quantum computing

Abstract doi.org/10.14569/IJACSA.2026.01708103 PDF