		FeedBack \| Contact Us \| Links \| Site Map

Welcome to Journal of Theoretical and Applied Information Technology

Submit Paper / Call for Papers

Journal receives papers in continuous flow and we will consider articles from a wide range of Information Technology disciplines encompassing the most basic research to the most innovative technologies. Please submit your papers electronically to our submission system at http://jatit.org/submit_paper.php in an MSWord, Pdf or compatible format so that they may be evaluated for publication in the upcoming issue. This journal uses a blinded review process; please remember to include all your personal identifiable information in the manuscript before submitting it for review, we will edit the necessary information at our side. Submissions to JATIT should be full research / review papers (properly indicated below main title).

Journal of Theoretical and Applied Information Technology
September 2026 | Vol. 104 No.17

Title:	SWIFTCACHE: AN RL-POWERED ADAPTIVE CACHING FRAMEWORK FOR LOW LATENCY CONTENT DELIVERY IN ECOMMERCE NETWORK
Author:	DR. P. MAHESWARI, Dr. V.SEEDHA DEVI, DR. AKILA VENKATRAMAN, KANCHARLA NAGABABU, DR. PERURI VENKATA ANUSHA, ELANGOVAN MUNIYANDY
Abstract:	Modern e-commerce networks experience rapidly changing product demand, geographically distributed requests, heterogeneous content, and freshness-sensitive information, creating challenges for existing reinforcement-learning-based caching approaches that are mainly designed for generic edge, CDN, or service-caching environments. This paper presents SwiftCache, an adaptive caching system powered by RL. It has been designed specifically to low-latency e-commerce content transmission and takes into account factors like network latency, content size, originality, cache state, and dynamic product popularity (as well as regional demand). A CDN/network simulation layer is utilised to account in edge nodes, geographic regions, cache capacity, heterogeneous objects, uniqueness, and network parameters. The RetailRocket E-Commerce Dataset and the M5 Walmart Dataset are utilised to model demand patterns and real user-product request behaviour, respectively, in the development of a trace-driven e-commerce CDN environment. Three current RL-based caching techniques are compared to SwiftCache under the same experimental conditions: EC-MADRL, Adaptive Layer-Wise Personalised Federated DRL, and Resource-Aware DRL. According to the simulated findings, SwiftCache attains 91.8% CHR, 90.7% BHR, 90.8% regional CHR, 48.6 ms P95 latency, and 3.1% FVR. SwiftCache increases BHR by 6.83–14.66%, lowers FVR by 54.41–64.37%, lowers P95 latency to 26.03–35.03%, and increases CHR by 6.62–11.41% in comparison to the three baselines. For refined and consistent low-latency content delivery, these results indicate that using demand, geography, content, and freshness factors that are specific to e-commerce in RL-based caching is better than incorporating generalised RL caching strategies.
Keywords:	E-Commerce Networks, SwiftCache, Network Latency, Cache Capacity, Request Behaviour
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	Q-ASCEND: A QUANTUM-RESILIENT ADAPTIVE CONSENSUS AND LIGHTWEIGHT VERIFICATION FRAMEWORK FOR CLOUD-SCALE BLOCKCHAINS
Author:	VELMURUGAN M, RAJEEV KUMAR M
Abstract:	The development of quantum computing is an inherent risk to the traditional blockchain security schemes based on RSA and elliptic curve cryptography. At the same time, deploying blockchains on clouds continues to suffer significant limitations in terms of high latency, high power consumption, scaling, and ineffective verification overheads. In this paper, I have introduced Q-ASCEND, which is a Quantum-Resilient Adaptive Consensus and Lightweight Verification Framework that is suitable in achieving secure, scalable, and energy efficient operation of blockchain technology under heterogeneous cloud-based settings. The framework has proposed lattice-based post-quantum cryptographic primitives, adaptive time-dependent consensus model, compact hash-linked verification model and energy conscious orchestration layer. Experimental evaluation was done in the dynamic workload condition with 50-1000 validator nodes. Findings prove that Q-ASCEND can provide a mean throughput of 198 transactions per second through 500 nodes and an average block confirmation latency of 184 ms. The energy used per block is cut to a maximum of 94.13% of Proof-of-Work models and the overhead of validation is lowered by an estimated 39%. Accuracy of consensus stability to churn conditions under node churn conditions is 99.4% at low churn and 77% at 50% churn. These results demonstrate that Q-ASCEND is a strong and future-herald blockchain system that can support high performance and quantum resilience future-generation cloud-native distributed systems.
Keywords:	Quantum-Resilient Blockchain, Adaptive Consensus, Post-Quantum Cryptography, Lattice-Based Signatures, Cloud-Scale Blockchain.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	DEVELOPMENT OF INFORMATION AND ANALYTICAL DECISION SUPPORT SYSTEMS IN GOVERNMENT RELATIONS FOR THE FORMATION OF VETERAN POLICY
Author:	LIUDMYLA ZUBRYTSKA, VOLODYMYR SERVETNYK, LYUDMYLA DROZACH, NATALIIA ISKHAKOVA, LYUDMYLA LYASOTA
Abstract:	The study addresses the lack of an integrated information-analytical architecture that links interagency data, decision-support analytics, and Government Relations (GR) communication in veteran policy. The aim was to design and scenario-verify a GR-oriented decision support system (DSS) and to compare the conditions of its application in the United States, Canada, and Ukraine. The research design combined cross-national comparative analysis of strategic and regulatory documents, functional-analytical modelling, a two-round Delphi procedure, Analytic Hierarchy Process (AHP) weighting, and exploratory correlation analysis; the expert panel comprised 42 specialists (14 per country). The United States obtained the highest Veteran Governance Decision Support (VG-DSS) index (0.89), Canada reached 0.84 and showed the strongest GR adaptability, whereas Ukraine had the lowest baseline VG-DSS value (0.58). The harmonisation (HG) index was 0.91 for the United States, 0.87 for Canada, and 0.64 for Ukraine. Scenario verification projected average increases of +0.03, +0.04, and +0.08, respectively, rather than effects of completed national implementation. Exploratory associations among DI, DG, TM, and GA ranged from r = 0.83 to 0.92; the comparatively weaker TM-GA association indicated that technological maturity alone does not fully determine communication adaptability. The research contribution is an integrated GR-DSS architecture that combines digital integration, data governance, technological maturity, GR adaptability, and policy harmonisation within one comparative assessment framework. Its practical implication is the use of the model as a pre-implementation diagnostic tool for prioritising interoperability, data standardisation, analytical functions, and stakeholder feedback mechanisms in public-sector digital transformation.
Keywords:	Veteran Policy, E-Governance, Local Self-Government, Public Service, Government Communications.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	MULTI-LANDMARK EARLY WARNING FOR NON-SUCCESSFUL COMPLETION IN STEM ONLINE COURSES: TEMPORAL CONSISTENCY, EXPLAINABLE PREDICTION, AND CAPACITY CONSTRAINTS
Author:	TINGTING XIAO, NURULLIZAM JAMIAT
Abstract:	Most online-learning early-warning studies rely on a single prediction point and rarely consider changes in the risk population, the accumulation of stage-specific information, and the match between alert volume and support capacity. Using the Open University Learning Analytics Dataset, this study developed a multi-landmark framework for predicting non-successful completion in STEM online courses, defined as Fail or Withdrawn. After retaining each learner’s earliest eligible record, 17,956 unique learners were included, with risk sets of 15,068, 14,411, and 13,906 learners at Days 30, 60, and 90. Learner assignments to training, validation, and test sets were fixed across landmarks. Logistic regression, random forest, and XGBoost performed similarly, with no model consistently superior across all landmarks and metrics. XGBoost was selected for interpretation and decision analysis because of its competitive performance and compatibility with TreeSHAP; test ROC-AUC values were 0.804, 0.851, and 0.876. SHAP results showed that assessment participation and stage-specific attainment became increasingly influential, while recent activity and persistence remained complementary signals. Validation-selected F2 thresholds achieved recall above 0.92 but alerted 81.4%, 74.4%, and 64.8% of test learners. Capacity-constrained analysis revealed a trade-off between earlier intervention time and later identification efficiency. The strongest predictive signals were not necessarily the earliest or most actionable, and high recall did not guarantee a manageable alert list. A multi-landmark early-warning system should therefore update the eligible population, available information, and decision threshold at each stage, rather than interpret later performance as strict longitudinal improvement.
Keywords:	Academic Risk Prediction; STEM Online Learning; Multi-Landmark Prediction; Explainable Learning Analytics; SHAP; Capacity-Constrained Intervention
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	AN INTELLIGENT AHP–MONTE CARLO DECISION FRAMEWORK FOR SUPPLY CHAIN RESILIENCE IN INDUSTRY 5.0
Author:	OUMAIMA ZAID, JABIR ARIF, FOUAD JAWAB
Abstract:	Intelligent decision-support strategies are increasingly needed to assess supply chain resilience under uncertain operating conditions. Resilience, in the era of Industry 5.0, is built on the interaction between human expertise, advanced technologies, and adaptive manufacturing capabilities. This article presents an intelligent decision-support framework that combines the Analytic Hierarchy Process (AHP) with Monte Carlo Simulation (MCS) to evaluate supply chain resilience under uncertainty. It embeds Human–Machine Collaboration (HMC) and Additive Manufacturing (AM) as critical Industry 5.0 enablers. AHP is used to prioritize resilience dimensions, whereas Monte Carlo Simulation assesses disruption frequency, recovery time, and service-level variability across multiple disruption scenarios. The results identify Flexibility and Human Oversight as the dominant resilience drivers, where Sustainability plays a moderate role and Machine Efficiency and ESG Alignment have lower sensitivity. Simulation also shows that the combination of HMC and AM helps enhance resilience and improve recovery performance under moderate and severe disruptions. The proposed computational and probabilistic decision-support framework supports resilience evaluation and strategic decision-making in Industry 5.0-enabled supply chains.
Keywords:	Decision Support, Industry 5.0, Supply Chain Resilience, Analytic Hierarchy Process, Monte Carlo Simulation.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	LOW-OVERHEAD, SERVER-ASSISTED TIME SYNCHRONIZATION MECHANISM FOR MQTT-SN–BASED IOT SENSOR NETWORKS
Author:	HOUSSEIN WEHBE, ALI MCHEIK, BILAL KANSO, LOREEN BAKERR
Abstract:	In many Internet of Things (IoT) systems, low-power sensors distributed in the environment continuously collect data from their surroundings and forward it to an application backend, where it is analyzed and acted upon. Including a timestamp in every reading helps keep events consistent and properly ordered, but requires the sensors and server clocks to be synchronized. This common challenge is usually solved by relying on frequent exchanges of dedicated timing messages between nodes, which increase network traffic and processing overhead on resource-constrained sensors. In this paper, we introduce a lightweight time synchronization mechanism for IoT deployments using MQTT-SN as the data delivery protocol. It leverages existing MQTT-SN messages to carry data timestamps to a central server, which applies an adaptive algorithm to identify possible clock desynchronization, limit the effect of transient delays, and send clock-adjustment notifications to sensors only when necessary. By minimizing dedicated synchronization exchanges and shifting computation to the server, the proposed mechanism reduces additional communication traffic and limits processing at the sensor. Simulation results in OMNeT++ show lower or more stable clock offsets than SNTP during the later part of representative runs. The simulations also show approximately 96–98% fewer synchronization messages than SNTP without increasing simulated sensor energy consumption. Under approximately matched conditions, the mechanism produced lower average clock offsets than those reported for an MQTT-based approach.
Keywords:	Time Synchronization, MQTT-SN, Server-Assisted, Wireless Sensor Networks, Low Overhead
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	FROM PROMPTS TO PROFICIENCY: HOW GENERATIVE AI SHAPES COMPUTING STUDENTS’ IT SKILLS
Author:	NUR AMLYA ABD MAJID, SUZIYANTI MARJUDI, MUHAMMAD FAIRUZ ABD RAUF, NUR ILYANA ISMARAU TAJUDDIN, ZURAIDY ADNAN, HASLIZA ABU HASSAN, HAZRI HAIDAR, MOHD FAHMI MOHAMAD AMRAN
Abstract:	As Generative Artificial Intelligence (GenAI) has grown at a rapid rate, the higher educational landscape has been dramatically altered. While GenAI offers immense opportunities for personalized, self-paced learning, its integration into computing curricula raises critical concerns regarding cognitive over-reliance, the potential erosion of fundamental technical skills, and the preservation of academic integrity. This study evaluates the impact of widespread GenAI adoption on the technical IT proficiency of 151 computing students at Malaysian private universities. This study is strictly delimited to undergraduate and postgraduate computing students, analyzing their usage frequency of major tools (including ChatGPT, Gemini, and DeepSeek) against key technical skills, academic engagement, and operational efficiency. Data was collected through a structured electronic survey, and quantitative analysis was executed within a Jupyter Notebook environment utilizing Cronbach's Alpha, One-Sample T-tests, ANOVA, and Multivariate Analysis of Variance (MANOVA) complemented by Tukey HSD post-hoc tests. The empirical results reveal that students maintain a moderate level of IT proficiency, demonstrating independent troubleshooting capabilities and foundational programming knowledge, alongside significantly enhanced academic engagement and operational efficiency. However, a significant portion of the cohort exhibited a critical dependency on AI-generated solutions, leading to a noticeable decline in independent debugging proficiency and logical problem-solving. Unregulated use of these tools threatens to bypass practice-based learning, leading to a phenomenon where academic performance is inflated while core technical skills are eroded. Based on these findings, we advocate for a "support-centered" pedagogical integration of AI that positions GenAI as an auxiliary learning partner rather than a total cognitive substitute, preserving hands-on problem-solving and critical thinking. This paper outlines clear practical implications for curriculum designers, discusses major threats to the study's validity, and provides future research directions, proposing nonparametric assessments like the Kruskal-Wallis H test and Spearman's Rank correlation to refine future educational insights.
Keywords:	Generative AI, IT Proficiency, Higher Education, Skill Erosion, Academic Integrity, Human-AI Collaboration
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	NEUROSYMBOLIC-HOMONYMSENSENET: INTEGRATING KNOWLEDGE GRAPH REASONING AND DYNAMIC SENSE MEMORY FOR EXPLAINABLE SEMANTIC REPRESENTATION LEARNING
Author:	MS. S. SUBI , DR. B. SHANTHINI
Abstract:	Semantic ambiguity caused by homonymous lexical forms remains a major challenge in Natural Language Processing because identical surface forms can represent unrelated meanings across contexts. Although transformer-based language models improve contextual representation, they may still underrepresent rare senses, lack explicit multi-hop semantic reasoning, and provide limited explanations for their predictions. This study introduces NeuroSymbolic-HomonymSenseNet, a unified explainable framework that combines a Context-Aware Transformer Encoder, Dynamic Sense Memory Bank (DSMB), WordNet–Wikidata knowledge graph reasoning, Graph Attention Reasoning Network (GARN), Neuro-Symbolic Semantic Inference Engine (NS-SIE), and Explainable Contrastive Multi-Sense Optimisation. The central contribution is the integration of contextual evidence, persistent sense prototypes, structured semantic relations, and symbolic inference within one representation-learning pipeline, with explicit mechanisms for rare-sense preservation and traceable decision support. The framework is evaluated on the SemCor and WiC benchmarks against Word2Vec, GloVe, BERT, RoBERTa, and SenseBERT. It achieves 97.83% accuracy, 97.46% precision, 97.21% recall, and 97.33% F1-score on SemCor, and 96.91% accuracy and 0.94 MCC on WiC. Rare-sense evaluation gives 95.18% recall and 95.42% F1-score, while explainability reaches 91.75%. Knowledge-graph evaluation reports 98.11% semantic consistency and 97.28% reasoning accuracy. These results indicate that combining neural contextual learning with memory-based sense preservation and symbolic graph reasoning can improve homonym disambiguation, rare-sense retention, and interpretability.
Keywords:	Neuro-Symbolic Learning, Homonym Disambiguation, Knowledge Graph Reasoning, Dynamic Sense Memory Bank, Explainable Artificial Intelligence, Semantic Representation Learning and Word Sense Disambiguation.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	DEVELOPING AUTONOMOUS CYBER DEFENCE SYSTEMS USING AI-DRIVEN INTRUSION DETECTION FOR IOT NETWORKS
Author:	BAHA ELDIN HAMOUDA
Abstract:	As the Internet of Things (IoT) continues to grow exponentially, securing interconnected devices against evolving cyber threats has become paramount. This paper explores the development of an autonomous cyber defence system leveraging artificial intelligence (AI) to enhance intrusion detection for IoT networks. Traditional security measures fall short in resource-constrained, heterogeneous environments like IoT. Hence, we propose an AI-driven Intrusion Detection System (IDS) capable of real-time threat detection and autonomous response. Using a hybrid machine learning approach and reinforcement learning for automated mitigation, the system is evaluated on publicly available IoT security datasets. Results demonstrate improved detection accuracy, reduced false alarms, and efficient resource utilization. This research contributes to the advancement of intelligent, self-governing security mechanisms in IoT ecosystems.
Keywords:	Autonomous Cyber Defence, AI-Driven Intrusion Detection, Internet of Things (IoT) Security, Machine Learning for Cybersecurity, Intelligent Network Protection.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	TOWARDS PRECISION ONCOLOGY: ENHANCED MULTISCALE DEEP FEATURE LEARNING FOR CERVICAL CELL SEGMENTATION
Author:	JHEELAM MONDAL , RAJDEEP CHATTERJEE , MAHENDRA KUMAR GOURISARIA
Abstract:	Over 300,000 people die from cervical cancer, which is the fourth most frequent malignancy in women worldwide. The early detection of cervical cancer corelates with significantly improved survival rates and the disease is largely preventable. It is indeed, a rare disease in affluent countries with robust immunization and screening initiatives. Nevertheless, the disease disproportionately impacts women in low- and middle-income nations, who often endure severe and untreatable diseases due to resource scarcity. Our objective is to create a reliable deep learning framework capable of distinguishing between normal and malignant morphologies by accurately delineating the cytoplasm and nucleus of the cells. We employ an innovative U-Net model named YOLOv5++ on a publicly accessible annotated dataset of cervical cytology images. Preliminary findings show the accuracy, IoU and Dice score values of 98.33%, 69.68% and 82.13% respectively which offers enhanced accuracy and efficiency compared to various other advanced U-Net methodologies. Thereby, our study demonstrates the enormous potential of deep learning-enabled segmentation as an effective technique to improve and standardize the cervical cancer screening process, reducing human error and enabling early clinical intervention.
Keywords:	Medical Image Analysis, Cervical Cancer, Pap Smear Images, U-Net Model, Image Segmentation
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	CAN WEIGHTED FEATURE ALLOCATION AND MULTI-LEVEL CLASSIFICATION IMPROVE VGG16-BASED ORAL CANCER CLASSIFICATION FROM DIGITAL IMAGES?
Author:	PATIBANDLA SANDHYA KRISHNA, SUBBA RAO PERAM
Abstract:	Oral squamous cell carcinoma (OSCC) is a serious health problem, and early detection of cancerous tissues is crucial. While deep learning has shown great promise in the field of automated oral cancer image analysis, current methods are increasingly becoming complex, such as those involving attention, transformer, ensemble, segmentation, and multimodal architectures, with comparatively less focus on explicitly emphasizing discriminative deep features prior to classification. In this study the gap is addressed by proposing a Weighted Feature Set with Multi-Level Classification using VGG16 (WFS-MLC-VGG) for automated oral cancer image classification. This framework first extracts deep representations from VGG16 and then performs feature selection, correlation-based feature weighting and multi-level classification to mitigate the impact of redundant features and highlight discriminative features. Experimental results show that the proposed model performs with an accuracy of 98.4% for feature-weight allocation and 98.8% for classification at the maximum number of images tested and the reported feature-processing times are smaller than those of the alternative models investigated for the current work. The findings offer evidence that while there is no need for adding more to the architecture of the CNN to get explicit feature prioritization, it can still be achieved by using explicit feature prioritization. The major contribution in the principal is a feature-management approach that combines weighted feature allocation and multi-level classification, as part of a diagnostic pipeline based on VGG16. The outcome, however, can only be applied to the dataset and experimental conditions used to evaluate the model, and clinical validation is required by external factors. The study thus confirms the methodological feasibility of the weighted feature processing for classification of oral cancer images and provides key directions in the way of developing robust, scalable and clinically validated computer aided diagnosis.
Keywords:	Oral Cancer, Feature Subset, Deep Learning, Convolution Neural Network, Histopathology Classification.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	ROBUST AND HIGH-PRECISION SPEED CONTROL OF BRUSHLESS DC MOTORS USING AN IMPROVED FAST TERMINAL SLIDING MODE CONTROLLER
Author:	CH. TRINAYANI, RAVI SRINIVAS LANKA
Abstract:	Brushless DC (BLDC) motors have gained widespread attention in industrial automation, robotics, electric vehicles, and precision drive systems because of their high efficiency, compact structure, and excellent dynamic characteristics. Despite these advantages, maintaining accurate speed regulation under varying operating conditions remains a challenging task due to parameter uncertainties, external disturbances, and sudden load variations. Conventional control techniques often exhibit limitations in terms of response speed, robustness, and tracking accuracy when subjected to such disturbances. To address these challenges, this study presents an Improved Fast Terminal Sliding Mode Control (IFTSMC) strategy for enhancing the speed control performance of BLDC motor drives. The proposed controller combines the robustness of sliding mode control with an improved terminal sliding surface design to achieve rapid convergence, reduced chattering effects, and improved steady-state behavior. A detailed mathematical model of the BLDC motor is developed, and the effectiveness of the proposed control scheme is investigated through comprehensive simulation studies conducted in the MATLAB/Simulink environment. The controller performance is examined under multiple operating scenarios involving changes in reference speed and load conditions. Furthermore, the proposed approach is compared with conventional Proportional–Integral (PI), Sliding Mode Control (SMC), Terminal Sliding Mode Control (TSMC), and Fast Terminal Sliding Mode Control (FTSMC) techniques. Simulation results demonstrate that the IFTSMC method provides faster transient response, smoother speed tracking, enhanced disturbance rejection capability, and significantly reduced steady-state error. In addition, the controller maintains stable operation and high control accuracy under dynamic operating conditions. The findings indicate that the proposed IFTSMC approach offers a reliable and efficient solution for advanced BLDC motor speed regulation, making it suitable for modern industrial applications that demand high precision, robustness, and superior dynamic performance.
Keywords:	Brushless DC Motor (BLDC), Speed Control, Improved Fast Terminal Sliding Mode Control (IFTSMC), Sliding Mode Control (SMC), Robust Control, Motor Drives, Dynamic Performance, Speed Tracking, Industrial Automation.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	QUANTUM PSO BASED TASK SCHEDULING ALGORITHM FOR MEDICAL APPLICATION IN CLOUD COMPUTING ENVIRONMENT
Author:	SUSHREE BHARATI, PRASANT KUMAR PATTNAIK, DIPTI DASH
Abstract:	Cloud computing has become a transformative technology in modern computing, allowing individuals and organizations to access software, hardware, and computing platforms remotely through the Internet. In the healthcare sector, it plays a crucial role in managing and processing large volumes of medical data and supporting complex healthcare applications. To ensure efficient operation in cloud environments, effective task scheduling algorithms are essential. These algorithms help reduce execution time and operational costs, improve resource utilization, maintain data privacy, and deliver a high Quality of Service (QoS). The findings of this study indicate that the Quantum Particle Swarm Optimization (QPSO)-based scheduling technique outperforms traditional scheduling methods. By combining intelligent and hybrid optimization strategies, QPSO achieves more efficient resource allocation and task execution. This approach is particularly beneficial in dynamic healthcare environments, where timely processing of medical data is critical. The results show that QPSO significantly improves response time, reduces data center processing time, and lowers overall operational costs, making it a promising solution for cloud-based healthcare systems.
Keywords:	Cloud computing, Comparative analysis, CloudSim, Load balancing, Makespan, Task Scheduling
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	EDGECROP: A LIGHTWEIGHT SQUEEZE-AND-EXCITATION-GUIDED ATTENTION CASE-NET FOR CROP WATER STRESS CLASSIFICATION ON EDGE MICROCONTROLLERS
Author:	DEEPIKA KATARIA , RAMESH KUMAR C
Abstract:	Autonomous irrigation decision-making at the field edge remains constrained by a persistent trade-off between classification accuracy and deployability on low-cost microcontrollers, particularly across the diverse agro-climatic conditions of Indian smallholder agriculture. This paper introduces EdgeCrop, a benchmark for crop water stress inference constructed from FAO-56 soil water balance simulation across multiple crops and agro-climatic zones, and evaluates sixteen candidate architectures spanning classical machine learning, convolutional, recurrent, and hybrid attention-based models. Motivated by the observation that no existing architecture jointly performs channel-level feature selection and temporal attention within a footprint suitable for microcontroller deployment, we propose CASE-Net, a hybrid architecture comprising a convolutional encoder, a squeeze-and-excitation recalibration block, and a self-attention encoder arranged sequentially so that channel recalibration conditions the features subsequently attended to by the attention mechanism. A full-capacity variant and a compressed variant, related through a capacity-transfer training procedure, are evaluated alongside the fifteen baseline models using classification accuracy, post-training quantization behaviour, simulated microcontroller latency and energy, and downstream irrigation-scheduling performance relative to a calendar-based baseline. Results show that the proposed architecture achieves leading classification accuracy at multiple compression tiers, with the compressed variant occupying a favourable accuracy-versus-footprint position relative to competing lightweight architectures and remaining within the energy and memory budgets of representative solar-powered microcontroller targets. A quantization anomaly specific to attention-based models is identified and explained. The scheduling simulation further reveals that classification accuracy alone is insufficient to predict irrigation-scheduling performance, underscoring the need to evaluate proposed architectures using both classification and downstream operational metrics rather than classification accuracy in isolation.
Keywords:	TinyML; Edge Computing; Crop Water Stress Inference; Irrigation Scheduling; Squeeze-And-Excitation Networks; Self-Attention; Knowledge Distillation; Capacity Transfer; FAO-56 Water Balance Model; Microcontroller Deployment; Precision Agriculture; Model Compression; Quantization; Smart Irrigation; Indian Agro-Climatic Zones
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	FORMATION OF AN INTEGRATED SYSTEM OF PUBLIC ADMINISTRATION IN THE FIELD OF NATIONAL SECURITY IN THE CONTEXT OF HYBRID AND INFORMATION THREATS
Author:	VOLODYMYR KYRYLENKO, LIUDMYLA LITVIN, IVAN TKACHENKO, SERHII HODLEVSKYI, KYRYLO TKACHENKO
Abstract:	The formation of an integrated digital public administration system in the field of national security in the context of hybrid and information threats is critically important for ensuring the stability of the state. The problem lies in the insufficient empirical validity of the combination of institutional indicators of governance quality (WGI) with an assessment of the implementation of information and computer technologies (GCI) for countries in a state of active hybrid war. The purpose of the study is to assess the institutional and technological potential of Ukraine to confront modern security challenges based on a comparative analysis with the countries of the European Union and the Baltics, as well as to determine the role of information and computer technologies in the development of an integrated security management model. The methodology is based on WGI and GCI data, using author's indices (INSGI, EUCI, CCI) to assess convergence. The results revealed a persistent institutional gap, but a significant strengthening of the digital component: GCI increased from 65.9 to 83.9 points, the level of cyber convergence reached 89%. A multi-level digital architecture of an integrated public administration system is proposed. The contribution of the study is the development and testing of a comprehensive INSGI index, the application of gap and convergence indices to quantify Ukraine's progress, and the justification of the strategic role of ICT in enhancing institutional capacity. The practical impact of the work is to create a basis for the implementation of digital solutions that allow accelerating convergence to European governance standards and increasing resilience to hybrid threats.
Keywords:	Information Technology, Computer Technologies, Digital Transformation, Public Administration, National Security
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	CONVOLUTIONAL BLOCK ATTENTION MODULE -ATTENTION SEGFORMER WITH UPERNET-STYLE MULTISCALE AGGREGATION (UMSFA) FOR POTHOLE DETECTION
Author:	S. SARASWATHI, SONIA JENIFER RAYEN
Abstract:	The roads are significant to trade and the development of economy yet they are in constant destruction by traffic and unfavourable weather conditions causing potholes that pose danger to people. The pothole detection is a critical aspect in automated road maintenance, self-driving vehicles, as well as in smart transportation systems. Detecting and risk assessment should be effective and this requires the potholes to be identified and the severity of the potholes be evaluated based on several factors including the size, depth, location, traffic density, and weather conditions. However, the current automated systems tend to give false positives, primarily in difficult situations due to shadows, stains or other interference of the environment. To overcome these issues, we suggest a novel deep learning model to detect the potholes and prioritize them based on the risk level, namely Convolutional Block Attention Module-Attention SegFormer (CBAMAS). This model combines the Mix Transformer (MiT) encoder and UPerNet-Style Multiscale Aggregation (UMSFA). The proposed MiT-CBAMAS model takes advantage of the hierarchical feature extraction of the data through the encoding based on the transformers and the multi-scale aggregation of contexts enhanced by attention mechanisms concentrating on the key spatial and structural features. The outputs of real-time detector are also further processed to categorize potholes based on the type of risk, and alert drivers to take immediate actions with these potholes. Our model was tested on pothole benchmark pothole-600. It had always performed better than the traditional transformer-based and CNN-based segmentation techniques. Having a Dice score of 94.36% and an IoU of 91.52, MiT-CBAMAS proposes edge sharp and segmentation consistency in problematic conditions in a suggestive manner. The results establish that the synergistic integration of UMSFA hierarchical decoding with MiT-CBAMAS greatly improves the model’s ability to detect irregular, low-contrast, and partially occluded potholes, providing a highly practical solution for real-time road damage assessment.
Keywords:	CBAM, SegFormer, Attention Mechanism, Pothole Detection, Semantic Segmentation, UPerNet, Transformer Backbone, Multi-Scale Feature Aggregation, MiT Encoder, Smart Transportation, Road Damage Analysis, Vision Transformers
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	ROBO-ADVISORY AS A TOOL FOR INVESTMENT CAPITAL MANAGEMENT IN DEVELOPED STOCK MARKETS
Author:	SVITLANA KOVALCHUK, DMYTRO NIKITIN, NATALIIA STELMAKH, ZORIANA LAPISHKO, VITALII FEDORYSHEN
Abstract:	Relevance: The rapid growth of robo-advisory necessitates the assessment of its effectiveness as a low-cost investment capital management tool in developed stock markets. Aim: The aim of the study is to determine the mechanism of influence of robo-advising on investment capital management and evaluate its effectiveness in developed stock markets. Methods: The research methodology combined risk-adjusted performance analysis, fixed effects panel regression, event-study, Difference-in-Differences, cluster analysis, and statistical robustness testing. This made it possible to evaluate robo-advisory not only in terms of profitability, but also in terms of risk, costs, stress resistance, institutional scaling conditions, and statistical reliability of the obtained results. Results: Robo-advising has proven its economic efficiency in developed stock markets (USA, UK, Germany, France, Switzerland, Japan, Singapore, and Australia). Annualized return was 7.8% versus 6.9% in traditional advisory portfolios and 7.2% in ETF benchmarks; Sharpe ratio reached 0.74, Sortino ratio – 1.06, Jensen’s alpha – +0.42 pp, max drawdown −18.4%. Its advantage was formed due to lower costs and portfolio discipline: fee −0.25 pp gave +0.31 pp net return, ETF share >80% – +0.54 pp, rebalancing reduced portfolio drift by 34.2%, tax-loss harvesting added 0.38–0.62 pp after-tax return. During the 2020 shock, subsidence was lower (−19.1% vs. −24.8%), recovery was faster (6.9 vs. 8.6 months), and the results remained statistically significant (t=2.87; p=0.006; U=418.0; p=0.011). Conclusions: The results showed that the effectiveness of robo-advisory advising was determined by a combination of risk-adjusted performance, lower costs, controlled portfolio drift, and more resilient investor behaviour during periods of market stress, which confirms the previously proposed hypothesis. The academic novelty of the study is the empirical proof of robo-advisory as an economic mechanism for capital management. Its effectiveness was formed through the interaction of return–risk–cost–tax–behavioural stability.
Keywords:	Automated Consulting, Wealth Management, Asset Allocation, Risk-Adjusted Performance, ETF Diversification, Portfolio Drift, Tax-Loss Harvesting
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	TRANSFORMER-BASED MULTI-MODEL FRAMEWORK FOR DUAL-TASK PREDICTION OF HEART DISEASE RISK WITH CLINICAL INTERPRETABILITY AND COMPUTATIONAL EFFICIENCY
Author:	KRISHNA KOMARAM, KHAZA K B VALI BHASHA SK, BODDUNA SRINIVAS, BANDI MADHUSUDHAN, VASAM RAMULU, KOTESWARA RAO KUMBHA, KALAM PARAMAIAH
Abstract:	Heart disease remains a leading cause of mortality worldwide, exerting severe physical and emotional impacts on affected individuals. Recent advances in deep learning have demonstrated remarkable potential in the diagnosis and prediction of cardiovascular conditions. This study proposes a novel framework based on a multi-model transformer architecture for predicting heart disease risk through both classification and regression tasks. Leveraging self-attention mechanisms, the model effectively captures complex interdependencies among clinical features such as age, cholesterol levels, and ECG signals. The architecture employs multiple branches to process categorical and continuous inputs, which are integrated using advanced attention layers to extract meaningful feature relationships. A key strength of the framework lies in its computational efficiency, achieved through parameter sharing, lightweight attention modules, pruning, quantization, and hardware-specific optimization—enabling deployment in real-world, resource-constrained healthcare systems. Experimental evaluation on benchmark datasets, including the Cleveland dataset from the UCI Machine Learning Repository, demonstrates superior performance compared to traditional machine learning models. The proposed model achieves high accuracy across classification and regression tasks, with attention mechanisms enhancing interpretability. Comprehensive performance metrics such as accuracy, precision, recall, F1-score, AUC-ROC, and AUC-PR confirm the model’s robustness. Overall, this transformer-based approach significantly advances the predictive landscape for heart disease, combining accuracy, interpretability, and practical efficiency.
Keywords:	Convolutional Neural Network (CNN), heart disease (HD), Cardio Vascular Disease (CVD), Interpolative layer perception model (IPLM), Intuitive Dedicated Deep-dense Model (IDDM).
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	ENTROPY STACKED VARIATIONAL DENOISING AUTOENCODER (ESVDAE) BASED DENOISING MODEL AND ATTENTION MULTI-SCALE DILATED CONVOLUTIONAL NEURAL NETWORK (AMSDCNN) FOR SUSPICIOUS-HUMAN-ACTIVITY RECOGNITION (SHAR)
Author:	LIKHITH SR, Dr. MAHALAKSHMI R
Abstract:	SHAR plays an important role in automated surveillance systems and security monitoring, where noise, cluttered backgrounds, and variations in activity scale can reduce recognition reliability. In this paper, we propose an end-to-end deep learning framework that incorporates an Entropy Stacked Variational Denoising Autoencoder (ESVDAE) and an Attention Multi-Scale Dilated Convolutional Neural Network (AMSDCNN) for the recognition of suspicious human activity. ESVDAE combines variational and denoising autoencoders through a stacked VDAE and entropy-guided representation learning to mitigate noise interference and preserve discriminative visual information. Afterward, the denoised images are used for ResNet50-based feature extraction, whereas AMSDCNN includes attention, dilated convolution, MSFF, and depthwise separable convolution. From an information technology standpoint, the suggested framework can serve as an integrative solution for noise-robust representation learning, multi-scale context modeling, and computation-efficient video classification. Experiments are conducted using reconstruction, recognition, discrimination, and computational metrics on the UCF-Crime and CUHK Avenue benchmarks. PSNR values are 41.32 dB and 43.89 dB, while SSIM values are 0.9785 and 0.9874 for UCF-Crime and Avenue, respectively. As for the activity recognition task, the accuracies obtained by AMSDCNN are 92.75% and 94.84%, and the AUC values are 92.87% and 94.69% on the two respective datasets. Furthermore, computations show 60 FPS inference with 12.68 million parameters. Overall, the results demonstrate the viability of the suggested framework as a means of SHAR. At the same time, its computational cost, inability to handle temporal data, and cross-dataset performance remain crucial aspects to study further.
Keywords:	Variational Denoising Autoencoder (VDAE), Entropy Stacked Variational Denoising Autoencoder (ESVDAE), Suspicious Human Activity Recognition (SHAR), Denoising, ResNet50,
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	ADAPTIVE DEEP MULTI-TIER ENSEMBLE FRAMEWORK WITH TEMPORAL FEATURE ENGINEERING AND GENERATIVE CLASS BALANCING FOR PRECISION STUDENT OUTCOME PREDICTION IN E-LEARNING ENVIRONMENTS
Author:	T MAHALAKSHMI , HEMANTH KUMAR VIROTHI , KARUNA ARAVA , MAKKAPATI SATYA SUKUMAR , S SINDHURA , VALLAM REDDY BHARGAVI REDDY
Abstract:	The task of student outcome prediction in e-learning platforms represents an extremely challenging high-dimensional analytics problem, being characterized by complex temporal dependencies, severe class imbalance, feature redundancy, and heterogeneous data modalities. None of the existing solutions are able to tackle all the aspects in an integrated and end-to-end predictive architecture that seamlessly integrates feature engineering, class balancing, model optimisation and explainability. The ADMEF (Adaptive Deep Multi-tier Ensemble Framework) is a new four-component pipeline that addresses each challenge individually, with complementary mechanisms working together. An LSTM-Autoencoder is proposed for temporal feature distillation, able to extract patterns of engagement trends from clickstream and activity logs; AF-3TS (Adaptive Filter-Wrapper-Embedded Three-Tier Selection) is presented as a cascaded feature selection architecture which progressively selects features via Relief-F filtering, RFE-Bayesian wrapping, and Elastic-Net embedded selection, ultimately reducing the feature set by 62.4% and losing just 1.3% predictive signal. A DCGAN-SMOTE model based on the conditional generative adversarial network (CGAN) coupled withk-means cluster-guided SMOTE interpolation is developed to balance class distributions, achieving a Frchet Inception Distance < 12.8; a QEL-4 (Quadrant Ensemble Learning) with a four-model stacking architecture is put forward which ensembles XGBoost, Bidirectional LSTM, TabNet with sequential attention, and CatBoost with Particle Swarm Optimisation, all combined by a Ridge-regularised meta-learner trained on out-of-fold probability vectors. Extensive experiments conducted on three standard e-learning datasets (OULAD, KEEL-Dropout and EduNet) indicate that ADMEF significantly outperforms the existing state-of-the-art regarding accuracy (0.9912 vs 0.958), F1-macro (0.9904 vs 0.9503), AUC (0.9989 vs 0.9867) and Matthews Correlation Coefficient (0.9871 vs 0.9448). Abolition studies have confirmed that every component has a statistically significant positive contribution (p < 0.01, Wilcoxon signed-rank test); Shapley values confirm that score trend, VLE click velocity and forum interaction are the primary factors. ADMEF provides real decision support to academic advisors and curriculum designers in identify at-risk students early, accurately and interpretably.
Keywords:	E-Learning Analytics, Student Dropout Prediction, Ensemble Learning, Feature Selection, Class Imbalance, LSTM, DCGAN, SHAP Explainability, Hyperparameter Optimization
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	BLOCKCHAIN-BASED DECENTRALIZED TRUST MANAGEMENT FOR 7G MULTI-VENDOR NETWORK SLICES
Author:	Dr.J.JAYAPAL, DR SWEETLIN SUSILABAI S, YAZHINI B, R.VANIDHASRI, S ANJALI DEVI, KURAGUNTLA RADHIKA, KAKUMANU ASHOK BABU, MAREESWARI G, Dr.T.VENGATESH
Abstract:	The evolution toward 7G networks introduces transformative capabilities alongside unprecedented challenges in trust management, particularly within multi-vendor network slicing environments. As telecommunications networks transition from monolithic, single-provider architectures to open, disaggregated ecosystems spanning multiple administrative domains, establishing and maintaining trust across heterogeneous stakeholders becomes paramount. This paper proposes a blockchain-based decentralized trust management framework tailored for 7G multi-vendor network slices. Leveraging permissioned distributed ledger technology and smart contracts, the framework enables automated service level agreement (SLA) negotiation, real-time monitoring, reputation-based provider selection, and immutable audit trails without centralized trust anchors. We present a comprehensive architecture integrating blockchain layers with network slicing orchestration, supported by a novel trust evaluation mechanism combining direct experience and third-party reputation. Experimental evaluation demonstrates the framework's effectiveness in ensuring trustworthy inter-provider agreements, with smart contract execution latency averaging 180 ms and successful detection of 95.6% of SLA violations. The proposed solution aligns with emerging 3GPP, ETSI, and ITU-T standardization efforts for decentralized trust in next-generation networks, offering a viable pathway toward secure, autonomous multi-domain 7G service orchestration.
Keywords:	Blockchain, Decentralized Trust Management, 7G Networks, Network Slicing, Multi-Vendor Environments, Service Level Agreements (SLA), Smart Contracts, Reputation Systems, Distributed Ledger Technology (DLT), Multi-Domain Orchestration, SLA Violation Detection, Permissioned Blockchain.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	AI-ADAPTIVE FEDERATED BLOCKCHAIN (AFBCHAIN): DYNAMIC RL-OPTIMIZED CONSENSUS AND XAI-ENHANCED FRAUD DETECTION FOR ULTRA-SECURE, SCALABLE DIGITAL PAYMENTS
Author:	VOREM KISHORE, SRIKANTH BHYRAPUNENI , DR. NARESH KUMAR BHAGAVATHAM , VENKATA BALA ANNAPURNA P, DR K LAKSHMAIAH5, MOHANA PRASAD M, DR. N. SATHEESH
Abstract:	Undoubtedly, there is a pressing need to create secure, scalable and decision-making payment systems in the age of rising cyber threats and boundless digital financial innovation. Given this fact, this paper introduces AFBChain, a state-of-the-art federated blockchain platform aimed at enhancing the efficiency and security of transactions and fraud detection powered by Hyperledger Fabric, reinforcement learning (RL), federated learning (FL), and explainable AI (XAI). AFBChain essentially operates a Proof-of-Intelligence (PoI) consensus mechanism and has opted to make the switch between PBFT and PoS and DPoS protocols into a Markov Decision Process (MDP), which is adept at adaptively switching choice according to the current network conditions to improve the through-put around 25 percent and decrease latency by 20 percent compared to Ethereum baselines. To ease fraud-detection and privacy-preserving results of above 95% accuracy, FL is inspired by deep-learning models, Gaussian Mixture Models, can be used to provide fraud-detection results, and SHAP and LIME models can be used to explain the constructed anomaly-detection knowledge. Also included in the framework will be dynamic sharding with the help of an artificial intelligence-based system, extreme audits of smart contracts to vulnerabilities of reentrancy, as well as decentralized automated market-making mechanisms in DeFi using RL by reducing slippage by 17.5%. The throughputs, scalability and security improvements are measured using the PAI-BC index. Overall, AFBChain has proven to be better than traditional blockchains, providing a stable and transparent system for the DeFi-based ecosystems, cross-chain payment and central bank digital currency.
Keywords:	AI-Blockchain Symbiosis; RL-Adaptive Consensus; Federated XAI Fraud; Dynamic Sharding Payments; DeFi Liquidity Optimization
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	QUANTUM-ENHANCED EYE-TRACKING BIOMARKERS TO IMPROVE EARLY DETECTION AND DIGITAL PHENOTYPING OF AUTISM SPECTRUM DISORDER: A HYBRID QUANTUM–CLASSICAL LEARNING APPROACH
Author:	SUNEETHA DAVULURI, ANJI REDDY TANGIRALA, VENUGOPAL BOPPANA, V P UMAMAHESWARA RAO MALLA, SREENIVAS VELAGAPUDI
Abstract:	Autism spectrum disorder (ASD) remains one of the foremost clinical challenges for early detection, and subjective diagnostic protocols constrain utilization while classical computational methods have failed to decode subtle oculomotor biomarker signatures. This work asks whether quantum feature encoding can move oculomotor ASD screening beyond the representational ceiling reached by classical eye-tracking classifiers, and answers this with a Hybrid Quantum–Classical Learning (QHCL) framework that combines parameterized quantum circuits with classical deep learning architectures to extract, encode, and classify eye-tracking biomarkers for early ASD detection and digital phenotyping. This is achieved by encoding raw gaze signals — fixation durations, saccadic velocities, scan path geometries, and pupillary response dynamics — into quantum Hilbert space representations through amplitude embedding to enable simultaneous processing of high-dimensional spatiotemporal oculomotor patterns. Quantum kernel-based feature learning reveals latent inter-feature correlations elusive to standard algorithms, and an entanglement-driven circuit layer forms enhanced patient-specific gaze embeddings. These embeddings are passed into a classical dual-head classifier trained for binary ASD classification and continuous digital phenotype evaluation along severity spectra. On three benchmark paediatric eye-tracking datasets, the proposed framework achieves classification accuracies of 95.8%, 97.4% and 96.2%, outperforming the strongest available classical baseline by 1.3–3.5 percentage points and improving quantum-versus-PCA feature learning by up to 9.3% at the individual feature-family level. The research contribution of this work is twofold: (i) it is the first study to apply quantum amplitude encoding and entanglement-driven parameterised circuits to oculomotor biomarker analysis for ASD, establishing that quantum Hilbert-space representations capture inter-feature correlations among fixation, saccadic, scanpath and pupillary signals that classical PCA-based pipelines discard; and (ii) it introduces a shared dual-head quantum-classical architecture that jointly optimises binary detection and four-class severity phenotyping from a single embedding, narrowing the accuracy gap that typically separates these two tasks in the literature. These findings open a new, non-invasive and computationally efficient pathway toward precision ASD screening with clinical applicability.
Keywords:	Autism Spectrum Disorder, Eye-Tracking Biomarkers, Hybrid Quantum-Classical Learning, Parameterised Quantum Circuits, Oculomotor Feature Encoding, Digital Phenotyping
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	SYSTEMATIC EVALUATION OF ROBUST LOSS FUNCTIONS FOR RESIDUAL AUTOENCODER-BASED SALT-AND-PEPPER NOISE REMOVAL IN BRAIN MRI
Author:	MANIKYA PRASUNA PAKALAPATI,HARISCHANDRA SIVA PRASAD NERSU, PRATHIMA TIRUMALAREDDY, SIRISHA ALAMANDA, CH VENKATESH, KAILASAM SWATHI
Abstract:	Salt-and-pepper noise (SPN) is a negative phenomenon in brain MRI, which causes deterioration of diagnostic quality by blurring the boundaries of tumors and disrupting downstream clinical processing. Although classical filters and deep learning autoencoders have been used to remove impulse noise, no systematic evaluation of robust loss functions to medical SPN denoising exists. It is the first systematic comparison of the Mean Squared Error (MSE), Huber, and Tukey bi-weight losses in a residual autoencoder architecture with symmetric encoder-decoder structure and multi-scale skip connections. We compared the performance on the BRISC 2025 and Alzheimer MRI datasets at different noise densities (10% -50%) and noise corruption types (SPN, Gaussian, and Rician noise and mixed noise) and assessed fidelity through PSNR and SSIM, perceptual quality through LPIPS and clinical applicability through Boundary F1 and downstream classification accuracy. As opposed to the theoretical expectations (of redescending property of Tukey), Huber loss (delta=1.0) at 30 per cent SPN (PSNR = 36.65 +1.86 dB, Boundary F1 = 0.868 +0.017) outperformed Tukey-based variants (DeltaPSNR =5.44dB, p=0.001). The results of our study indicate that gradient continuity is more advantageous than aggressive outlier suppression in suppressing high-density impulse noise in medical images. The ResAE-Huber model lies on the Pareto-optimal frontier of accuracy-efficiency trade-offs, providing tumor edge continuity at a cost of 4.83 ms inference latency and just 1.86M parameters. These findings form Huber loss as the objective of choice in residual autoencoder-based SPN reduction, and practical guidelines on clinical implementation with edge preservation and inference latency being paramount.
Keywords:	Medical Image Denoising, Residual Autoencoder, Robust Loss Functions, Salt-And-Pepper Noise, Tumor Boundary Preservation, Huber Loss, Tukey's Bi-Weight.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	FEDERATED EDGE LEARNING FOR ENERGY-EFFICIENT AND PRIVACY-AWARE ANOMALY DETECTION IN SMART GRID IOT INFRASTRUCTURES
Author:	DR. RAJAT VERMA, C N RAJALAKSHMI, DR. G. K. JABASH SAMUEL, ⁠DR. NEELAKANTHA GURU, KISHORE BALASUBRAMANIAN, DR T.B SIVAKUMAR, A V PRABU
Abstract:	Smart-grid operators rely on the deployment of smart meters and remote terminal units (RTUs) as part of the Internet of Things (IoT). Centralised conventional machine-learning detectors require to ingest the customers data, drive communication and energy costs, add latency, and create a single point of failure. Existing FL detectors in this setting require extensive usage of homomorphic encryption, do not include energy control mechanisms that are appropriate for low-power edge nodes or only provide detection results in simulation. In this paper, we introduce FEL-AD-SG, a federated edge-learning framework for anomaly detection, which is based on differential privacy and secure aggregation, model quantization, energy-aware client selection and duty cycling. Raspberry Pi 4 class nodes serve as the local training and inference nodes. The framework is tested in a reproducible simulation with Flower and TensorFlow, in a non-IID partition and with a hardware prototype, with the Ausgrid smart-meter corpus, a synthetic RTU trace based on Schneider Electric PowerLogic T300 parameters, and with standard intrusion-detection benchmarks. On Ausgrid, FEL-AD-SG achieves 98.5% accuracy and 97.5% F1-SCORE, which is on par with or better than a centralised baseline, and 59% less memory, 65% less CPU, 42% less energy, and 83% less communication compared to a Paillier-based federated detector. The study provides a complete experimental setup, a clear privacy–accuracy–energy trade-off for smart-grid IoT nodes with limited resources, a conceptual model and hypotheses, as well as a hardware resource profile.
Keywords:	Federated Edge Learning, Anomaly Detection, IoT for Smart Grid, Energy Efficiency, Raspberry Pi Prototype
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	A HYBRID QUANTUM-CLASSICAL FRAMEWORK FOR PNEUMONIA DETECTION FROM CHEST X-RAY IMAGES USING EFFICIENTNET-B4 AND DATA RE-UPLOADING VARIATIONAL QUANTUM CIRCUITS
Author:	V. R. NIVEDITHA, SASHI KANTH BETHA, DHRUVA R. RINKU, VASU GERA4, RISHTI5, TANISHA GOYAL, RENY JOSE, D.VETRITHANGAM
Abstract:	This paper presents a hybrid quantum-classical framework for binary pneumonia classification from chest X-ray images. A fine-tuned EfficientNet-B4 convolutional backbone serves as a fixed feature extractor, producing 1,792-dimensional representations that are compressed through a classical bridge network to eight scalar values. These values are subsequently processed by an 8-qubit Variational Quantum Circuit (VQC) that employs the data re-uploading strategy, encoding input features before each of three variational layers, for a total of 48 learnable quantum parameters. All quantum operations are performed using the PennyLane Lightning Qubit simulator; no physical quantum hardware was used, and the results therefore do not constitute a demonstration of quantum advantage over classical methods. The model is evaluated on the Kaggle Chest X-Ray (Pneumonia) benchmark (5,856 images) using a single train/validation/test split (5,016/200/624 images, with the 200-image validation set sampled from the training data); no cross-validation, repeated training runs, or statistical significance testing were performed, and the reported figures should accordingly be read as a single-run result rather than as estimates with associated confidence intervals. At a classification threshold of t = 0.59 selected on the test set itself, and therefore subject to optimistic bias, the model achieves 93.75% accuracy, 93.98% precision, 96.15% recall, 95.06% F1-score, and 0.9729 ROC-AUC. A two-phase training procedure a frozen-backbone stage followed by selective layer unfreezing, together with a feature pre-extraction cache- was used to make simulation-based quantum training practical on a single consumer GPU; both measures are reported for reproducibility rather than as research contributions in themselves. EigenCAM visualizations are provided as a qualitative illustration of model attention and suggest, without independent radiologist verification, that the model tends to attend to lung parenchyma regions broadly consistent with radiological pneumonia indicators; these interpretability observations should be treated as preliminary.
Keywords:	Pneumonia detection; Hybrid quantum-classical neural networks; Variational quantum circuits; EfficientNet-B4; Data re-uploading; Quantum simulation; Chest X-ray classification.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	TOWARDS THE DESIGN OF QUALITY FRAMEWORK OF KNOWLEDGE MANAGEMENT SYSTEM EFFECTIVENESS: AN EMPIRICAL APPROACH
Author:	SIVANESH KUMAR A, ERANA VEERAPPA DINESH S, GOMATHY NAYAGAM M, SOMASUNDARAM K
Abstract:	Globalization has made Knowledge Management Systems (KMS) indispensable for organizations seeking competitive advantages in today's interconnected economy. However, achieving effectiveness in KMS remains a significant challenge due to its interdisciplinary nature, encompassing software engineering, information technology, cognitive science, marketing, and business management. The complexity stems from the need to simultaneously address quality aspects (system and process) and success aspects (organization, users, and benefits). This study investigates KMS effectiveness across five dimensions: system quality, process quality, organizational support, user satisfaction, and perceived benefits. Data were collected from 452 knowledge workers across six sectors in India (nuclear power, software development, services, manufacturing, healthcare, and education) and analyzed using Partial Least Squares (PLS) structural equation modeling. The findings reveal that organizational support has the strongest association with system quality (β=0.55), while process quality demonstrates high predictive strength (R²=0.843), and user satisfaction shows the strongest overall explanatory power (R²=0.883) among all dimensions. Notably, the hypothesis linking user satisfaction to organizational performance was rejected (β=0.00), indicating that user satisfaction alone does not directly translate to organizational performance gains a finding that challenges conventional IS success models. The study provides a validated roadmap for KMS developers, practitioners, and researchers, offering prioritized quality dimensions and measurable indicators (KQIs) for assessing KMS effectiveness. The proposed Unified Quality Model (UQM) bridges critical gaps in understanding KMS effectiveness and establishes empirically-grounded guidelines for implementation and evaluation
Keywords:	Knowledge Management Systems (KMS), KMS Effectiveness, System Quality, Process Quality, Organizational Support, User Satisfaction, Perceived Benefits, Unified Quality Model (UQM), Partial Least Squares (PLS), India.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	NEUROCARDIOAI: AN AI-POWERED FRAMEWORK INTEGRATING EEG AND COGNITIVE MARKERS FOR CARDIAC RISK PREDICTION
Author:	MAGANTI GOUTHAM,DR. BHAVANA JAMALPUR
Abstract:	Cardiovascular diseases (CVDs) are also one of the most important causes of death internationally, making risk prediction essential for early detection. The traditional paradigm heavily demarcates physiology-based markers, i.e., ECG, BP, and lipid profiles, while overlooking all the neurological and cognitive dimensions that cardiometabolic health can affect. While brainwave patterns, mental performance, and cardiac risk factors all have significant and clinically relevant associations individually and have recently been incorporated, these modalities have rarely been incorporated into a unified diagnostic paradigm. Existing models are mostly unimodal and use shallow or domain-specific machine learning techniques that do not capture inter-modality relations and temporal dynamics in a complex fashion. To address these limitations, we propose a novel AI-powered, multi-modal deep learning framework for early cardiac risk prediction: NeuroCardioAI. Leveraging EEG, cognitive test performance, and cardiovascular indices, the system enables modeling of multimodal health states. At the heart is BrainWaveHeartNet, a new hybrid CNN-Transformer architecture that extracts spatial patterns from EEG and sequential semantics from cognitive markers and fuses them with cardiovascular data at various levels using an attention-based fusion layer. We showed through extensive experiments on publicly available EEG and cardiovascular datasets that AUC-ROC of 98.63%, accuracy of 96.89%, precision of 96.40%, and recall of 97.10% can be achieved, significantly outperforming baseline models Interpretable Representation Learning for Empathetic Neural Conversation Models: Code for Healthy or Unhealthy Emotion Detection Methods (i.e. SVM, Random Forest, XGBoost). The modality gains are additive to performance, which is further validated with an ablation study. The model here has a scalable, non-invasive, real-time approach that is best suited to clinical (and memoir-based) preventative healthcare.) A major step in the direction of the convergence of neurocognition and cardiac-related diagnostic solutions, NeuroCardioAI is at least a step towards holistic, patient-centered, and proactive mitigation of cardiovascular risk.
Keywords:	Cardiac Risk Prediction, EEG Signals, Cognitive Markers, Deep Learning, Multi-Modal Fusion
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	ANCHORED IDENTITY, MOBILE BALLOT: A HYBRID CRYPTOGRAPHIC VOTING ARCHITECTURE FOR SAME-STATE, SAME-CONSTITUENCY DISPLACED VOTERS
Author:	M. MOHANA DEEPTHI, MS. D V V DEEPTHI, DR B MALATHI, KARUTURI KAVYA RAMYA SREE, KOLAKALURI ANURADHA, MS. ANURADHA P, DR. N. SATHEESH
Abstract:	In every general election including state elections, there is a huge mass of registered electors who are unable to reach the polling place to which they are placed. But a disqualification is rarely caused by legality, rather by geographical factors. Approximately 400 million/more citizens lose the ability to vote because of inter-state labor, intra-state assignment, hospital admission, caring for the sick, and traveling to hospital. In the current work an idea to extend the trusted electronic voting machine paradigm rather than to replace it is introduced to present an One India / One State Distributed Hybrid Voting System (OIDHVS), which is cryptographically secure and supervised remote voting system. Its structure includes a national digital layer to verify a voter using the KYC parameters, and create a home constituency ballot; a tokenization layer to convert the marked ballot into an Encrypted Vote Token and use an additively homomorphic encryption method, zero-knowledge proof of ballot validity and threshold decryption to verify and deposit the same; and a physical layer of supervised nodes in administrative offices, transit centers, and hospitals to verify and deposit a token in the form of a tamper evident paper artefact at the home constituency. The major contribution that this paper makes is the Same-State Same-Constituency Displaced Voter mechanism which addresses the hitherto unmet elector who even though they are on the roll call of a particular constituency, they happen to be in a different location within the same state on the day of voting. Such votes are captured separately on a dedicated separately audited sub-counter on the single-constituency tally, rather than the proposed disenfranchisement or return trip, which is impractical. With performing the simulation through the sharding mechanism, pocketing to the next Indian general election that is made in the year 2024, more than ninety-two per cent of voter sessions under a five-second service guarantee, exception for voters' authentication to be stood below two per cent and throughput of sharded validators is at least proportional to national demand, the simulation has been validated on the national level with published hardware, connectivity and biometric protocols. These results are not to be interpreted as evidence of technical viability and suitability of the architecture to meet its expected throughput, latency and reliability criteria within the provided assumptions, but can only be interpreted as evidence of technical feasibility (showing how the architecture could work to meet these criteria), not legal and/or security and/or readiness for deployment (which would require prototype implementation, independent of legal and security certification, formal legal analysis and field piloting).
Keywords:	Remote Voting; Encrypted Vote Token; Homomorphic Tallying; Zero-Knowledge Proof; Voter Authentication; Displaced Voter
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	META-DSS: A SMART DECISION SUPPORT SYSTEM FRAMEWORK FOR SUSTAINABLE MANGROVE ECOTOURISM MANAGEMENT
Author:	DARWIN YUWONO RIYANTO, RUDI SANTOSO, JANUAR WIBOWO, MOCHAMMAD ARIFIN
Abstract:	Mangrove ecotourism has emerged as a strategic approach to achieving environmental conservation, community empowerment, and sustainable coastal economic development. However, the management of mangrove tourism destinations in Indonesia continues to face significant challenges, including fragmented stakeholder coordination, limited digital integration, inadequate tourism information management, and the absence of systematic decision-support mechanisms. These limitations hinder the ability of tourism managers and policymakers to make data-driven decisions that balance ecological sustainability, visitor experience, and local economic growth. This study aims to develop META-DSS (Mangrove Ecosystem Transformation Approach–Decision Support System), a smart decision support system framework designed to support sustainable mangrove ecotourism management through the integration of digital technologies, stakeholder collaboration, and sustainability indicators. This research employed a qualitative exploratory approach involving in-depth interviews, field observations, document analysis, and focus group discussions with tourism managers, local communities, MSMEs, government representatives, academics, and tourists at selected mangrove ecotourism destinations in East Java, Indonesia. The collected data was analyzed using thematic analysis, stakeholder mapping, ecosystem analysis, and framework synthesis techniques to identify critical components required for an integrated tourism decision support system. The findings reveal that sustainable mangrove ecotourism management requires the integration of five key components: stakeholder collaboration, tourism information systems, digital marketing and visitor engagement, environmental sustainability monitoring, and community-based economic development. Based on these findings, the study proposes the META-DSS framework consisting of a stakeholder layer, data acquisition layer, analytics layer, decision support layer, and sustainability outcomes layer. The framework enables systematic information flow, real-time tourism monitoring, evidence-based decision-making, and collaborative governance among tourism stakeholders. The study concludes that META-DSS provides a comprehensive smart tourism framework capable of enhancing destination management effectiveness, strengthening sustainability performance, supporting local economic participation, and facilitating digital transformation in mangrove ecotourism ecosystems. The proposed framework contributes to the development of tourism decision support systems and smart sustainable tourism management in emerging coastal destinations.
Keywords:	Decision Support System; Smart Tourism Ecosystem; Mangrove Ecotourism; Digital Transformation; Sustainable Tourism Management
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	CONDITIONAL TABULAR GENERATIVE ADVERSARIAL NETWORK WITH RECURRENT ATTENTION NETWORK FOR DATA IMBALANCE IN MEDICAL DATA
Author:	JAYANTHI MUDDAPPA, JAYASUDHA KALIANNAN, M. JAGADEESAN
Abstract:	The class imbalance problem is critical in medical diagnosis due to large-scale clinical datasets exhibiting an imbalanced class distribution. Data balancing algorithms are employed to address this problem by oversampling minority class or undersampling majority class. In this manuscript, Conditional Tabular Generative Adversarial Network (CT-GAN) is developed in preprocessing phase to balance the data and improve classification performance. CT-GAN learns the feature distributions and generates samples that preserve feature correlation, which enhances the classification model’s performance. In the classification phase, the Recurrent Attention Network (RAN) is developed to classify the performance of balanced data instances in diabetes. The RAN integrates a Recurrent Neural Network (RNN) with an attention mechanism for focusing more on essential features of data. The developed CT-GAN with RAN obtains 97.85% accuracy on the PIMA dataset, 99.75% accuracy on heart dataset, and 99.00% accuracy on Statlog dataset.
Keywords:	Class Imbalance, Conditional Tabular Generative Adversarial Network, Feature Correlation, Feature Distributions, And Recurrent Attention Network.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	DEEP GRAPH ATTENTION NETWORKS FOR MRI BRAIN TUMOR CLASSIFICATION AND SPECTRAL ANALYSIS
Author:	RAJESWARI R , SUJATHA N
Abstract:	Classification of brain tumors based on MRI is essential in neuro-oncology. Conventional CNN algorithms treat images as structured grids and do not necessarily consider spatial interaction between image regions. In our paper, we develop a method that represents an MRI image through a graph structure, treating image regions as graph nodes associated with feature vectors and modeling interactions between regions through weighted edges. We experimentally compare GCNs with the proposed GAT model, which incorporates edge features into the attention mechanism. We evaluate their ability to classify three classes of tumors, namely glioma, meningioma, and pituitary tumors, achieving a classification accuracy of 77.37%, which surpasses the performance of GCN (71.53%) and conventional approaches (76.67%). Our spectral analysis of graph node features shows that different structures characterize each tumor type. Moreover, the proposed attention mechanism allows us to visualize diagnostically valuable areas of the image.
Keywords:	MRI, Brain Tumor Classification, Graph Neural Networks, Graph Attention Networks, Spectral Graph Properties
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	N-NET: POBI-LSTM: ENHANCED U-NET ARCHITECTURE WITH DUAL ENCODER MODEL FOR CT IMAGE SEGMENTATION WITH PARAMETER OPTIMIZED BI-LSTM
Author:	SATHYA SUNDARAM M, KARTHICK S, THIYAGARAJAN P
Abstract:	Intracranial hemorrhage (ICH) is a condition where blood flows into the brain due to trauma or medical disorders, requiring rapid medical and surgiacal care due to its high mortality rate and potential life-threatening consequences. Accurate and automated organ segmentation methods are crucial for diagnosis and treatment planning. Previous research on ICH segmentation using U-Net models has shown inadequate feature transfer rates, high model size, and slower speed. This research proposes a parameter-optimized version of N-Net, called N-NET: POBi-LSTM, to diagnose ICH images of the human brain. The model uses Bi-Directional Long Short-Term Memory (Bi-LSTM) for classification, with hyperparameters optimized using Fuzzy Elite Opposition Social Spider Optimisation (FEOSSO). The research emphasizes the importance of assisting the encoder in extracting more detailed characteristics and proposes an N-Net for enhanced segmentation of CT images. The dual encoder model is proposed to increase the depth of the U-Net network and improve feature extraction capabilities. The method incorporates the Squeeze-and-Excitation (SE) module for channel-level global characteristics and full-scale skip links for hybrid of higher and lower-level details. The model is compared to UNet, UNet++, and Dense-Net201_U-Net in terms of quantitative assessment using DSC, IoU, Accuracy, and loss evaluation.
Keywords:	U-Net, FEOSSO, Bi-LSTM, CT image, Segmentation
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	DYNAMICALLY STACKED ENDOSCOPY IMAGE QUALITY ENHANCED CAPSULE DEEP BELIEF NETWORK FOR GASTROINTESTINAL TRACT DETECTION
Author:	K. SHARIFA , S. MALARVIZHI
Abstract:	The growing pervasiveness of gastrointestinal (GI) tract disorders globally heightens the intense requirement for accurate diagnosis, as these diseases considerably influence human life and bring about high mortality rates. An effective treatment mechanism is indispensable for addressing this critical health issue. Early diagnosis with accurate treatment of gastrointestinal diseases is crucial for reducing mortality and boosting quality of life. In this study, a novel Dynamically Stacked Capsule Deep Belief Network (DS-CDBN) is proposed for comprehensive gastrointestinal disease detection. They are image quality enhancement and classification for gastrointestinal disease detection. First, an image quality enhancement model is applied to different categories of the Kvasir dataset by applying distance vector field and Dynamically Stacked Median Filter not only preserves the edge but also dynamically applies median filters to detect noise candidates and process accordingly. Next, rotation-based data augmentation is performed with varied magnitudes aids in detecting the disease for different categories. Finally, Capsule Proximal Greedy Layer-wise Deep Belief Network-based Gastrointestinal Tract Disease Detection is performed. Initially, features are learnt using a vector activation function called squashing through a capsule network. Then, to the learnt features proximal policy greedy layer is applied to generate diagnostic predictions for gastrointestinal tract diseases in an accurate and precise manner. Experimental evaluation on the publicly available Kvasir dataset achieved 27% accuracy and significantly lower training time by 37% compared to state-of-the-art methods. The results confirm the method’s efficiency for accurate classification of eight gastrointestinal tract diseases, offering a scalable and interpretable solution for enhanced clinical decision-making in gastroenterology.
Keywords:	Gastrointestinal, Dynamically Stacked Median Filter, Capsule Proximal Greedy Layer-wise, Deep Belief Network.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	MODIFIED FLEXIBLE NODE AUTHENTICATION FOR MANETS: HASH CHALLENGE–RESPONSE FOR SECURE DYNAMIC SOURCE ROUTING (DSR)
Author:	AHMAD K. ALOMARI
Abstract:	Mobile Ad Hoc Networks (MANETs) are a novel technology that has a dynamic topology, self-configuring, and self-motivated. The primary feature of an ad hoc network isn’t dependent on any established infrastructure because it lacks a centralized arbiter or server. Wireless MANETs have several issues, including performance analysis, energy efficiency, security, and network stability. To date, a great deal of research has been conducted to create security strategies for MANETs. The current strategies that aim to build a defense against different types of attacks at different levels will be covered in this analysis. Researchers therefore create several routing protocols. Wireless networks cannot effectively employ the routing methods designed for conventional networks. A few novel routing methods have been developed for wireless ad hoc networks that are appropriate for the constantly evolving ad hoc wireless environment. In this study, we proposed a technique to improve the efficiency of routing protocols and boost node-to-node dependability in mobile ad hoc networks (MANETs). Our system focuses on node authentication, and we use an on-demand routing protocol, such as the Dynamic Source Routing protocol (DSR), to implement it. This technique relies on the hash function, secret value, and time stamp.
Keywords:	MANETs, Routing protocols, hash function, DSR, source node, destination node
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text

Title:	WCE-SECUREAI: A LIGHTWEIGHT HYBRID CRYPTOGRAPHIC FRAMEWORK FOR SECURE DATA TRANSMISSION IN IOMT USING WIRELESS CAPSULE ENDOSCOPY
Author:	SHRAVANI AMAR , P.ANITHA
Abstract:	Wireless Capsule Endoscopy (WCE) is a diagnostic tool used for the detection of gastrointestinal (GI) diseases due to its non-invasive approach to visualizing the digestive tract. Nonetheless, the application of DPA in Internet of Medical Things (IoMT) is limited due to twofold difficulties, including secure medical data transmission and high computational complexity, as well as the poor applicability of current cryptographic schemes to resource-constrained devices. Standard solutions like AES-CBC with RSA suffer from high latency and significant payload expansion. At the same time, pure symmetric schemes lack effective key management and are also unable to provide adaptive protection. This highlights the necessity of a lightweight, scalable architecture that provides high-performance and secure WCE-based IoMT settings. In this work, we present WCE-Secure AI, a holistic framework architecture with a newly proposed hybrid encryption module (LightHybridCrypt) that utilizes AES-GCM for authenticated encryption and elliptic curve cryptography (ECC) for efficient key exchange. The KeyGuard-ECC protocol provides secure session management, while the EntropyAwareEncrypt submodule utilizes an adaptive strategy based on entropy only to harden encryption in high-entropy domains. The algorithm presented provides confidentiality and integrity while minimizing the latency and statistical attacks on it. We validate WCE-SecureAI over the Kvasir-Capsule dataset, demonstrating that it achieves up to 40% reduction in encryption time, 55% reduction in key exchange latency, ≤8% payload overhead, and up to 7.99/8 bits of entropy, with 99.1% integrity verification accuracy. The results demonstrate that the framework offers an optimal trade-off between efficiency and security, which makes it well-suited for real-time deployment in IoMT. Proposed System: The secure and robust data transfer of WCE through our proposed system lays the foundation for eventual integration with automated diagnostic pipelines into clinical practice.
Keywords:	Wireless Capsule Endoscopy, Internet Of Medical Things, Hybrid Encryption, Secure Data Transmission, Lightweight Cryptography
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th September 2026 -- Vol. 104. No. 17-- 2026
Full Text