		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
August 2026 | Vol. 104 No.16

Title:	CLINXAI-NET: A RULE-GUIDED HYBRID CNN–TRANSFORMER EXPLAINABLE AI FRAMEWORK FOR ROBUST REAL-TIME DIAGNOSIS IN HIGH-RESOLUTION MEDICAL IMAGING
Author:	NAGA NIRMALA KONGARA, DR. BHUKYA KRISHNA, DR. B. SUJATHA
Abstract:	High-resolution medical imaging supports increasingly precise diagnosis, but recent CNN, Transformer, and hybrid pipelines often optimize predictive discrimination separately from clinical reasoning, explanation faithfulness, calibration, and deployment cost. This fragmentation limits trustworthy use across modalities and motivates the present study. ClinXAI-Net is proposed as a rule-guided hybrid CNN-Transformer framework that combines lesion-sensitive local convolutional features with global anatomical Transformer representations through adaptive gated fusion. Clinical IF-THEN consistency constraints and an explanation-alignment objective are incorporated into training and decision refinement, while Grad-CAM, SHAP, LIME, and textual rule traces provide complementary evidence. Five public datasets covering MRI, CT, PET, and X-ray were evaluated with patient-wise splits, five-fold validation, discrimination metrics, expected calibration error, quantitative explanation criteria, and deployment measures. The framework achieved 99.12% test accuracy, 98.82% F1-score, and 0.994 ROC-AUC. In matched ablation analysis, the full model improved accuracy by 2.20 percentage points and explanation agreement by 0.100 over CNN-Transformer fusion alone. INT8 quantization reduced model size by 40.6%, inference latency by 29.2%, and peak memory by 23.5%, with only a 0.06 percentage-point reduction in accuracy relative to the non-quantized full model. The contribution is therefore not merely another hybrid backbone; the study provides evidence that jointly optimizing predictive, rule-consistency, explanation, calibration, and deployment objectives can improve diagnostic performance while preserving auditable reasoning. The findings support a more clinically accountable design pattern for medical-image decision support, while prospective multi-center validation remains necessary before clinical adoption.
Keywords:	Explainable Medical Imaging, CNN-Transformer Fusion, Clinical Rule Reasoning, Explanation Alignment, Real-Time Diagnosis
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	FROM DIAGNOSTIC AUTOMATION TO TRUSTWORTHY CLINICAL INTELLIGENCE: A SYSTEMATIC SCOPING REVIEW AND EVIDENCE-TO-DEPLOYMENT ROADMAP FOR ARTIFICIAL INTELLIGENCE IN DIGITAL DENTISTRY
Author:	GUMMA PARVATHI DEVI, DR. PRATHIPATI RATNA KUMAR
Abstract:	Digital dentistry increasingly reports high-performing computational models, yet a central knowledge gap remains: existing reviews often catalogue technologies or deployment concerns without a common framework that distinguishes analytical performance from transportability, clinical validity, clinical utility, and lifecycle safety. Consequently, it remains unclear what level of real-world use is actually justified by a reported benchmark result. This systematic scoping review critically appraised literature available through July 2026; 147 records were screened and 56 studies were included for evidence synthesis. Rather than pooling incomparable accuracy values, the review evaluated data provenance, partitioning and leakage, reference standards, external validation, calibration and uncertainty, fairness, human factors, cybersecurity, prospective impact, and lifecycle monitoring across imaging, natural-language processing, multimodal and foundation models, robotics, AR/VR, tele-operation, synthetic data, and digital twins. The synthesis identifies an accuracy-to-deployment gap: evidence is comparatively mature for selected two-dimensional imaging and segmentation tasks, whereas multimodal assistants, generative systems, robotics, tele-operation, and digital twins remain limited by external, prospective, human-factor, or lifecycle evidence. The novelty of this review is an integrated evidence-to-deployment taxonomy, an M0-M5 clinical utility ladder, minimum reporting requirements, and staged evidence gates. The new knowledge created is that evidence maturity, not nominal accuracy alone, determines the defensible level of clinical use, and that the validation burden increases as autonomy, multimodality, generative behaviour, and post-deployment change increase. These outputs provide a practical basis for designing, reviewing, and governing clinically credible digital-dentistry research.
Keywords:	Artificial Intelligence; Digital Dentistry; Clinical Translation; Foundation Models; Multimodal Learning; Explainable AI; Calibration; External Validation; Robotics; Lifecycle Governance.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	ENHANCING SECURITY AND ENERGY EFFICIENCY IN 5G HETNETS USING DEEP REINFORCEMENT LEARNING
Author:	SAI PRASANTH KANUPARTHY, MEDESWARA RAO KONDAMUDI, LAVURI SANKAR, S V KIRANMAYI SRIDHARA, ANUSHA BASAMSETTI, DR. K.B. GLORY, DR. SARALA PATCHALA, GARAGA SRILAKSHMI
Abstract:	The paper presents a new method known as SecBoost. It has proposed to enhance the security aspect and energy consumption of the 5G network system. To achieve this, we employ deep reinforcement learning (DRL) in order to establish an optimised self-adaptive intelligent system that learns and optimises itself in response to the current conditions in real-time. In today’s multi-level systems, various base stations offer access through the use of millimetre wave (mmWave) signals. These signals are however fast but they are prone to cracking by the benedict who may try to hack through the data. These are tough nut to crack especially using the traditional approaches to security because they compromise energy efficiency. To address issues the paper presents a learning-based solution called SecBoost to combat the discussed problem. SECBoost is developed through reinforcement learning, a kind of artificial intelligence tool that can make right decisions concerning the security and functionality of the network. In this, implemented a multi-agent reinforcement learning (MARL) approach. Power control determines the amount of power used for transmission, channel allocation selects the best frequency channels and beamforming directs signals to specific users. To handle the large and complex decision-making process, use a dueling deep Q-network (D3QN). SecBoost performs better than traditional methods. It is a strong candidate for future wireless security solutions. It provides both security and energy efficiency. This makes it useful for next-generation communication systems.
Keywords:	SECBoost, Reinforcement learning model, HetNets, Channel selection, Beamforming.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	A KPI-ALIGNED ADAPTIVE AI–INFORMATION SYSTEM ARCHITECTURE FOR ENTERPRISE DECISION SUPPORT
Author:	SAWSAN .G. SAMMOUR, FADWA ISSA AHMAD ALSALIM, MAHER MOHAMMAD ALNAIM, HASSAN ALI AL-ABABNEH, MAHER IBRAHIM TAWDROUS, JAMEEL AHMAD KHADER
Abstract:	Enterprise AI projects tend to be loosely coupled with enterprise information systems and business key performance indicators (KPIs) and affect individual analytical tasks. This study tackles this enterprise decision support IT design problem by creating an adaptive AI–information system architecture aligned with the corresponding KPIs. Following the design-science research (DSR) approach, information-systems, AI-capability, AI-governance, and machine-learning-lifecycle literature was used to define the requirements, which are then instantiated in three interacting layers: 1) data integration, 2) analytics and optimization, and 3) KPI feedback and governance. Ex ante evaluation of the artifact was done in five cross-industry process scenarios: forecasting, allocating resources, decision latency, financial utility, and adaptation stability. The numbers given are normalization of scenario outputs and not the treatment effects of the companies involved. This evaluation reveals that the important value of the architecture lies in the end-to-end traceability of data, model output, optimizations and deviations in KPIs. The scenarios predict gains for the accuracy of forecasting of approximately 15–18%, reduction in time for decision making of up to 30%, and resource efficiencies of around 10–15% under the stated assumptions, but these gains diminish when the quality, interoperability and governance requirements for the data are not met. The IT research contribution is a reusable reference architecture and feedback-control logic to transform the disjointed AI applications into an auditable enterprise decision-support lifecycle. Practical implications are staged integration with ERP/CRM and supply-chain systems, human approval gates and ongoing drift, security and unintended KPI checks.
Keywords:	Artificial Intelligence; Enterprise Information Systems; Decision Support; Design Science; Adaptive Architecture; KPI Alignment
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	AGRIDA-SSL: A DOMAIN-ADAPTIVE SELF-SUPERVISED LEARNING FRAMEWORK FOR ROBUST MULTI-CROP AGRICULTURAL IMAGE ANALYSIS ACROSS DIVERSE ENVIRONMENTS AND GROWTH STAGES
Author:	MODALAVALASA DIVYA, Dr. BHUKYA KRISHNA, Dr. CH. RAMESH
Abstract:	Precision agriculture requires image-analysis models that remain reliable when crop type, phenological stage, environment, and sensing platform change, yet recent supervised and self-supervised studies commonly address a single image source or a single downstream task. This study presents AgriDA-SSL, a domain-adaptive self-supervised framework that learns from unlabeled satellite, UAV, and field-level imagery and is fine-tuned with limited annotations. The framework combines a hybrid CNN-transformer encoder with contrastive representation learning, masked image reconstruction, prototype-guided clustering, domain-consistency regularization, and an adaptive augmentation policy. The evaluation covers crop classification, disease recognition, field segmentation, and yield-related visual estimation across PlantVillage, BigEarthNet/Sentinel-2, Agriculture-Vision, LUCAS crop images, and UAV imagery. Under the reported experimental protocol, AgriDA-SSL attains 98.91% accuracy, 98.64% precision, 98.52% recall, 98.58% F1-score, and 0.993 ROC-AUC for the principal classification experiment, while ablation and low-label analyses indicate that each major component contributes to performance. The study's contribution is therefore not a claim that self-supervision alone is new, but the integration and evaluation of complementary self-supervised objectives, cross-domain regularization, and adaptive augmentation within one multi-source, multi-task agricultural vision pipeline. The results support label-efficient agricultural monitoring, while the reported threats to validity delimit the extent to which benchmark performance can be generalized to unseen farms, sensors, seasons, and deployment conditions.Precision agriculture requires image-analysis models that remain reliable when crop type, phenological stage, environment, and sensing platform change, yet recent supervised and self-supervised studies commonly address a single image source or a single downstream task. This study presents AgriDA-SSL, a domain-adaptive self-supervised framework that learns from unlabeled satellite, UAV, and field-level imagery and is fine-tuned with limited annotations. The framework combines a hybrid CNN-transformer encoder with contrastive representation learning, masked image reconstruction, prototype-guided clustering, domain-consistency regularization, and an adaptive augmentation policy. The evaluation covers crop classification, disease recognition, field segmentation, and yield-related visual estimation across PlantVillage, BigEarthNet/Sentinel-2, Agriculture-Vision, LUCAS crop images, and UAV imagery. Under the reported experimental protocol, AgriDA-SSL attains 98.91% accuracy, 98.64% precision, 98.52% recall, 98.58% F1-score, and 0.993 ROC-AUC for the principal classification experiment, while ablation and low-label analyses indicate that each major component contributes to performance. The study's contribution is therefore not a claim that self-supervision alone is new, but the integration and evaluation of complementary self-supervised objectives, cross-domain regularization, and adaptive augmentation within one multi-source, multi-task agricultural vision pipeline. The results support label-efficient agricultural monitoring, while the reported threats to validity delimit the extent to which benchmark performance can be generalized to unseen farms, sensors, seasons, and deployment conditions.Precision agriculture requires image-analysis models that remain reliable when crop type, phenological stage, environment, and sensing platform change, yet recent supervised and self-supervised studies commonly address a single image source or a single downstream task. This study presents AgriDA-SSL, a domain-adaptive self-supervised framework that learns from unlabeled satellite, UAV, and field-level imagery and is fine-tuned with limited annotations. The framework combines a hybrid CNN-transformer encoder with contrastive representation learning, masked image reconstruction, prototype-guided clustering, domain-consistency regularization, and an adaptive augmentation policy. The evaluation covers crop classification, disease recognition, field segmentation, and yield-related visual estimation across PlantVillage, BigEarthNet/Sentinel-2, Agriculture-Vision, LUCAS crop images, and UAV imagery. Under the reported experimental protocol, AgriDA-SSL attains 98.91% accuracy, 98.64% precision, 98.52% recall, 98.58% F1-score, and 0.993 ROC-AUC for the principal classification experiment, while ablation and low-label analyses indicate that each major component contributes to performance. The study's contribution is therefore not a claim that self-supervision alone is new, but the integration and evaluation of complementary self-supervised objectives, cross-domain regularization, and adaptive augmentation within one multi-source, multi-task agricultural vision pipeline. The results support label-efficient agricultural monitoring, while the reported threats to validity delimit the extent to which benchmark performance can be generalized to unseen farms, sensors, seasons, and deployment conditions.Precision agriculture requires image-analysis models that remain reliable when crop type, phenological stage, environment, and sensing platform change, yet recent supervised and self-supervised studies commonly address a single image source or a single downstream task. This study presents AgriDA-SSL, a domain-adaptive self-supervised framework that learns from unlabeled satellite, UAV, and field-level imagery and is fine-tuned with limited annotations. The framework combines a hybrid CNN-transformer encoder with contrastive representation learning, masked image reconstruction, prototype-guided clustering, domain-consistency regularization, and an adaptive augmentation policy. The evaluation covers crop classification, disease recognition, field segmentation, and yield-related visual estimation across PlantVillage, BigEarthNet/Sentinel-2, Agriculture-Vision, LUCAS crop images, and UAV imagery. Under the reported experimental protocol, AgriDA-SSL attains 98.91% accuracy, 98.64% precision, 98.52% recall, 98.58% F1-score, and 0.993 ROC-AUC for the principal classification experiment, while ablation and low-label analyses indicate that each major component contributes to performance. The study's contribution is therefore not a claim that self-supervision alone is new, but the integration and evaluation of complementary self-supervised objectives, cross-domain regularization, and adaptive augmentation within one multi-source, multi-task agricultural vision pipeline. The results support label-efficient agricultural monitoring, while the reported threats to validity delimit the extent to which benchmark performance can be generalized to unseen farms, sensors, seasons, and deployment conditions.
Keywords:	Self-Supervised Learning, Precision Agriculture, Domain Adaptation, Crop Classification, UAV Imagery
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	BIG DATA ANALYTICS AND ARTIFICIAL INTELLIGENCE IN DIGITAL MARKETING PERFORMANCE MANAGEMENT
Author:	HASSAN ALI AL-ABABNEH, ASMAA KANAAN, RAGHAD AL MUHAREB, HAZIM HADDAD, MAJD AL MUHAREB, HADEEL MARZOUQ TBEISHAT, JAMEEL AHMAD KHADER
Abstract:	The management problem addressed in this study is that fragmented digital-marketing analytics cannot jointly process heterogeneous high-volume data, adapt predictions to changing customer behavior, explain model recommendations, and optimize financial and engagement outcomes. This limitation reduces managerial trust and makes ROI-oriented decisions difficult in omnichannel environments. The study proposes a Hybrid Explainable Artificial Intelligence and Big Data Architecture (HEAIBDA) that integrates distributed data processing, Random Forest, XGBoost, Long Short-Term Memory models, SHAP/LIME explanations, and a composite Digital Marketing Performance Index (DMPI). An explanatory, longitudinal, comparative multiple-case design is used with harmonized secondary indicators for ten technology-oriented corporations during 2020–2025. The evaluation compares a 2020–2022 baseline with a 2023–2025 HEAIBDA-aligned optimization scenario. Mean ROI increases from 17.20% to 26.13%, conversion rate from 3.31% to 4.88%, customer engagement from 0.659 to 0.832, and DMPI from 0.628 to 0.818, while customer acquisition cost decreases from USD 37.00 to USD 25.10. Prediction transparency rises from 47.66% to 83.61%, and all ten cases move in the expected direction. The study concludes that combining scalable analytics, hybrid prediction, and explainability offers a coherent decision-support architecture for multidimensional marketing performance management. Because the evidence is based on secondary aggregate data and scenario evaluation rather than controlled field deployment, the findings demonstrate design plausibility and cross-case consistency, not causal impact.
Keywords:	Big Data Analytics; Artificial Intelligence; Digital Marketing; Machine Learning; Explainable AI; Predictive Analytics; Marketing Performance Management; ROI Optimization
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	FEDERATED INTRUSION DETECTION IN IOT ENVIRONMENTS USING HGS-CS-BASED FEATURE SELECTION AND A GATED TRANSFORMER-BILSTM ARCHITECTURE WITH SWARM AGGREGATION
Author:	P. UMA DEVI, GURPREET SINGH CHHABRA
Abstract:	Introduction The increasing complexity of cyber threats in Internet of Things (IoT) environments requires the development of efficient and privacy preserving Intrusion Detection Systems (IDS). IoT networks generate high-dimensional and heterogeneous traffic data, making effective feature selection and temporal attack detection challenging, particularly in distributed and resource-constrained environments. Objectives: This research has the following objectives: (1) to develop a hybrid feature selection method using Hybrid Grey Wolf Optimization integrated with Cuckoo Search and chaotic Lévy flight (HGS-CS), (2) to reduce the dimensionality of IoT network traffic while retaining highly discriminative features, (3) to develop a TSTformer-LSTM model combining a Transformer Encoder, Bidirectional LSTM, and Gated Fusion mechanism for capturing global contextual and temporal dependencies, and (4) to develop a privacy-preserving federated IDS using SwarmFed with attention-based aggregation, Stable Focal Loss, and centralized fine-tuning. Methods: This research proposed a novel framework of hybrid feature selection using Hybrid Grey Wolf Optimization integrated with Cuckoo Search enhanced with chaotic Levy flight (HGS-CS) and a TSTformer-LSTM model integrated with a Transformer Encoder with a Bidirectional LSTM and a Gated Fusion mechanism that captures both the global dependencies in context and also the temporal dependencies which validated on the datasets UNSW-NB15 and BoTNeTIoT-L01. The HGS-CS performs the multi-objective feature selection which results in a dimensionality reduction of 86.96% by selecting only 3 highly discriminative features in BoTNeTIoT-L01 dataset and 8 features in UNSWNB-15 dataset. The model is trained based on a SwarmFed federated learning framework with attention-based aggregation, Stable Focal Loss and centralized fine-tuning. Results: Experimental results show performance which achieves an accuracy of 96.9% on BoTNeTIoT-L01 dataset and 89% on UNSWNB-15 dataset. The proposed framework provides a robust, efficient, and privacy-aware IDS solution suitable for distributed and resource-constrained IoT environments
Keywords:	Transformer Encoder, Bidirectional LSTM, UNSW-NB15, BoTNeTIoT-L01, Federated learning.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	A LIGHTWEIGHT HYBRID QUANTUM CLASSICAL FEDERATED LEARNING MODEL FOR ANOMALY DETECTION IN EDGE IOT NETWORKS
Author:	V T VENKATESWARLU , JILLELLA VENKATESWARA RAO , SELVA MALAR.N , KANDRAKUNTA CHINNAIAH , A. SRINIVASA REDDY⁵, BOSUBABU SONGA , PARUCHURI VENKATA KRISHNAKANTH , MORSA CHAITANYA
Abstract:	The number of IoT devices continues to rise and it is imperative to detect anomalies in real-time to ensure the security, reliability and efficiency of the system. Current deep neural network models have great memory and computing requirements, so it is not easy to use on the edge device. The benefit of federated learning is that it leaves raw data on the individual device, thus improving privacy. Nonetheless, efficient feature learning by compressing the size of quantum datasets has not been built into current federated learning frameworks, and the efficacy of quantum-based approaches under realistic noise in the hardware have not been widely investigated. To address privacy concerns, QEdgeNet is a lightweight hybrid quantum-classical federated learning (QLFL) model for anomaly detection in edge-IoT networks . In this model, the features of network traffic are first compressed by principal component analysis, and then the network traffic is mapped to a compact 4-qubit variational quantum circuit for the classification of anomalous traffic. This model is an aggregated model updated using the Cheon-Kim-Kim-Song (CKKS) homomorphic encryption mechanism, but without exposing the raw network data of clients, and raw traffic data never escapes from a client device during the training process. The applicability of the suggested approach in an idealized simulator is quantified by measuring the inference performance in a simulator simulated quantum hardware noise, the accuracy loss from the hardware noise is recovered by extrapolating to zero noise.
Keywords:	Edge-IoT, Quantum Machine Learning, Hybrid Quantum–Classical Learning, Federated Learning, Anomaly Detection, Variational Quantum Circuit (VQC)
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	SMARTEDGE SCHEDULER: A LEARNING-BASED FRAMEWORK FOR INTELLIGENT TASK SCHEDULING IN EDGE-CLOUD SYSTEMS
Author:	D. VENKATESWARLU, DR. BNV MADHU BABU, K. SWATHI, DR. V. SANGEETHA, DR. P. RAMESH
Abstract:	Despite the development of some efficient edge-cloud resource scheduling schemes, the efficient task scheduling of these schemes remains an unsolved problem, as existing schemes address only a subset of the scheduling challenge: heuristic, meta-heuristic, and deep reinforcement learning. A heuristic method can be light on computation but is inflexible for static processing. The deep reinforcement learning (DRL) schedulers adjust to the system's dynamics, yet they suffer from unpredictable convergence and a lack of predictive capability. Though metaheuristic optimisers improve search efficiency, they are incapable of learning reusable policies. This is the first framework to place these three strategies under a single adaptable architecture. Hereby, this paper contributes 4 original findings to fill this gap. It brings in two new advancements; firstly, it proposes a new hybrid task scheduling framework: SmartEdgeScheduler, which is a co-design of deep reinforcement learning, LSTM-based predictive offloading and metaheuristic optimisation, a combination previously unattempted in the literature. Second, it suggests an LSTM-based Predictive Offloading Model (POM) that predicts network latency and task execution time for future time slots, enabling proactive task placement before deterioration in the system is noticeable, shifting the status quo from reactive correction to anticipatory control. Thirdly, it provides an adaptive mode switching mechanism that is driven by a real-time DRL error metric, which activates the metaheuristic optimisation method when reinforcement learning convergence fails to be reached, thereby ensuring that it does not impair performance when reinforcement learning convergence is stable while the mode switching is needed when the reinforcement learning convergence is unstable. Fourth, it offers a Multi-Criteria Decision-Making (MCDM) module that brings explicit (and therefore auditable) task-priority information – missing in policy-implicit DRL schedulers – into the problem, based on (predicted) latency, resource utilisation efficiency, and energy cost. We have evaluated our SmartEdgeScheduler against a number of baseline and state-of-the-art algorithms on 10k task instances, showing: task completion time reductions of 25%, resource utilisation improvements of 15%, energy savings of 12%, and SLA compliance of 97%. In scaled testing with 5,000 to 20,000 tasks, controlled performance degradations are observed, demonstrating viability in a production environment. The demonstration of the need to co-design proactive prediction, adaptive policy switching, and multi-objective optimisation – rather than applying them separately – for reliable, energy-efficient, and scalable scheduling in heterogeneous edge-cloud systems is new knowledge generated by this work.
Keywords:	Task Scheduling, Edge-Cloud Computing, Deep Reinforcement Learning, Predictive Offloading, Metaheuristic Optimization
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	INTERNAL FRAUD DETECTION IN FINANCIAL INSTITUTIONS USING MACHINE LEARNING TECHNIQUES
Author:	JOSUE CORREA , SOLANGE ROJAS
Abstract:	This research aimed to implement a machine learning model aligned with Sustainable Development Goal No. 12: Responsible Consumption and Production, which focuses on strengthening access to information and technological knowledge. The objective of this study was to develop a machine learning model to enhance the detection of internal fraud within a financial institution located in Sullana, Piura, Perú. This was an applied research study, employing a quantitative approach, a pre-experimental design, and a descriptive scope. The population consisted of 6000 operational records from the financial institution and 25 employees from the business department. Observation guides and questionnaires were used as data collection instruments during the execution of the project. The results showed a 37% reduction in unusual behaviors, a 50-minute decrease in the average detection time, a 27.5% increase in accuracy, and a 96% rate of high employee perception. These findings indicate that the machine learning model proved to be an effective tool for fostering a culture of transparency, thereby reducing operational risks and reinforcing trust in the local financial system.
Keywords:	Risk Management, Financial Institutions, Artificial Intelligence, Data Processing, Automation
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	HYBRID CNN–LSTM MODELS FOR REAL TIME CHEMICAL PLUME RECOGNITION USING DRONE MOUNTED ELECTRONIC NOSE SYSTEMS
Author:	RAVURI DANIEL, BODE PRASAD2, A.DIVYA, KANDRAKUNTA CHINNAIAH, Dr. SHAHEDA NILOUFER, PARUCHURI JAYASRI, KADIYALA SUDHAKAR, RANJEET ASHOK KADU
Abstract:	Real-time identification of hazardous chemical plumes in complex outdoor environments remains a critical challenge in environmental monitoring, industrial safety, and disaster-response operations due to turbulent dispersion dynamics, sensor drift, and airflow disturbances induced by unmanned aerial vehicles (UAVs). This work proposes a hybrid CNN–LSTM framework for intelligent chemical plume recognition using a drone-mounted, bio-inspired electronic nose (E-nose) system designed for ppm-level detection of ammonia (NH₃) and methane (CH₄) gases under dynamic atmospheric conditions. The proposed platform integrates a multi-sensor fusion architecture comprising metal-oxide semiconductor (MOS), electrochemical, and environmental sensing units to capture heterogeneous spatiotemporal gas signatures during UAV-assisted monitoring missions. To mitigate the adverse effects of rotor downwash and plume fragmentation, an artificial intelligence (AI)-driven denoising and drift-compensation is introduced, incorporating adaptive baseline correction, temporal smoothing, and feature normalization to enhance signal stability during aerial navigation. The core recognition engine combines convolutional neural networks (CNNs) for local spatial feature extraction with long short-term memory (LSTM) networks for temporal dependency modeling, enabling robust discrimination of transient plume patterns in highly noisy environments. Experimental evaluations conducted across simulated industrial leakage scenarios demonstrate that the proposed hybrid architecture achieves superior plume classification performance, with an average accuracy exceeding 96%, while maintaining low inference latency suitable for real-time edge AI deployment on embedded UAV computing platforms. Furthermore, the framework exhibits enhanced robustness against sensor degradation, environmental turbulence, and concentration variability compared with conventional machine learning approaches. The proposed system establishes a scalable and autonomous aerial olfaction framework for next-generation environmental surveillance, hazardous gas leak localization, and intelligent atmospheric sensing applications using AI-enabled UAV ecosystems.
Keywords:	UAV gas sensing, Electronic nose, CNN–LSTM, Chemical plume recognition, Edge AI.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	EFFICIENT AND SCALABLE FRAMEWORK METHODS FOR DIMENSIONALITY REDUCTION IN HIGH-DIMENSIONAL BIG DATA
Author:	T BALAJI, K PRASUNA, HAROON RASHEED, SATTI RAMA GOPALA REDDY, ANJANEYULU NAIK R, PRAVEEN TUMULURU, N JAYA
Abstract:	The fast growth of high-dimensional big data has really made a mess of data storage, processing, visualization, and predictive analytics. In many cases, you end up with a huge number of features, and that tends to bring redundancy, noise, and extra computational burden. it can easily mess with the efficiency and the final behavior of machine learning models. This work puts forward an “Efficient and Scalable” Framework for dimensionality reduction in high-dimensional big data analytics, built on an Autoencoder (AE) approach. The idea uses a neural network structure so it can learn a compact yet meaningful representation from the original high-dimensional input. At the same time, it should keep the most important underlying information. Compared to traditional linear dimensionality reduction methods, this Autoencoder model is able to capture more intricate nonlinear connections between features, so it fits better for large-scale and mixed-source datasets. The framework, more or less, covers data pre-processing, feature normalization, Autoencoder-based representation learning, latent-space dimensionality reduction, and then downstream analytical assessment. After that the learned lower-dimensional embedding are used for classification, and prediction work, so we can judge how effective they are in practice. The evaluation uses key performance measures such as reconstruction error, accuracy, precision, recall, F1-score, computational time, and the dimensionality reduction ratio. Finally, results are contrasted with well-known dimensionality reduction techniques, to see whether there is a real improvement in information preservation, computational efficiency, and predictive performance. The Autoencoder architecture can be tweaked some more, mostly by choosing a suitable hidden-layer setup, picking the right activation functions, tuning the learning rate and batch size, plus using regularization methods that make things more robust. The aim is to boost both scalability and overall generalization, and not just on paper. In the experimental outcomes, we’d expect the method to noticeably shrink the dimensionality, and also ease the computational burden on those high-dimensional datasets while still keeping the discriminative signals needed for reliable analytics. Overall, this framework is flexible and can scale with large volume data, so it can be applied to things like healthcare, finance, cybersecurity, IoT, and other domains where data is heavy and constant. So, the proposed autoencoder based dimensionality reduction idea really acts as a strong alternative to traditional dimensionality reduction approaches, and it also becomes a base layer for building efficient big data analytics systems
Keywords:	Dimensionality, LDA, ICA, PCA,AE, Reduction
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	IMPROVING IMAGE FUSION QUALITY WITH APC-FUSENET: AN AUTOENCODER AND PCNN-BASED METHOD
Author:	BATTIREDDY SURENDARA , S. ABIRAMI , GURRAM SUNITHA
Abstract:	The objective of image fusion is to develop a single composite image that effectively incorporates the most relevant and informative details from each image respectively. In this research, a new multi-focus image fusion method, APC-FuseNet model is proposed. The model integrates an autoencoder with a Pulse-Coupled Neural Network (PCNN) to improve the quality and focus of the fused images. The methodology commences with loading and preprocessing of Multi Spectral (MS) and Panchromatic (PAN) images to ensure compatibility. Subsequently, Bilateral and Laplacian of Gaussian (LoG) filters are applied to the PAN image to preserve edges and suppress noise. The PAN image is then separated into detail, coarse, and base layers. A pre-trained autoencoder is used to extract compressed features. The Multi-Scale Morphological Gradient (MSMG) is utilized to enhance prominent characteristics, while a PCNN analyzes the MSMG outcomes to generate binary maps for the purpose of merging selective region. The ultimate merged image is acquired by integrating these outcomes into the HSV color space and then converting it back to Red Green Blue (RGB). The APC-FuseNet model's effectiveness is demonstrated through visual inspection and metric analysis, thereby revealing its superior performance in ERGAS (2.30564), UIQI (0.99521), CC (0.99634), PSNR (44.012), and SSIM (0.92017) compared to other methods. This confirms its ability to generate high-quality, focused fused images.
Keywords:	Autoencoder, Deep Learning, Multi Focus, Panchromatic Satellite, Pulse-Coupled Neural Network, Remote Sensing.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	HALO: A FUZZY MEMBERSHIP FRAMEWORK FOR MULTI COMMUNITY NODE ASSIGNMENT IN NETWORKS
Author:	HICHAM SADIKI, RAJAE ZRIAA, SAID AMALI
Abstract:	Leader-based community detection methods build communities around influential leader nodes rather than treating every node symmetrically. The recent algorithm TALB fuses topological similarity and node-attribute similarity into a single information matrix, but like its predecessors, commits every node to exactly one hard community, a poor fit for networks in which nodes belong to more than one social or thematic group. This study contributes new knowledge on the generalisability of fuzzy-membership overlap detection across leader-based and non-leader-based backbones and on the previously unquantified gap between synthetic and real-world overlap-detection accuracy. We propose HALO, a fuzzy-membership mechanism that reuses TALB's own information matrix, at no extra computational cost, to turn any hard partition into an overlapping one via a tunable threshold parameter tau. We test the hypothesis that HALO's benefit is attributable to the mechanism itself rather than to a specific backbone. HALO-TALB, applied directly on TALB, improves the Omega index and F1 score by up to 9 points over hard TALB on a synthetic benchmark, though a topology-only baseline remains competitive at high noise. HALO-GM, which decouples the mechanism from TALB's leader heuristic using a greedy-modularity backbone, resolves this gap, raising the Omega index from 0.79 to 0.99 and confirming that the mechanism generalizes across backbones. However, on ten real ego-Facebook networks with genuine overlapping ground truth, HALO never improves on a plain hard partition, even under oracle threshold selection; we trace this to the much richer overlap structure of real social circles and identify adapting HALO's membership rule to this regime as the key open problem. These findings support the mechanism's backbone-independence but show its practical benefit depend on overlap density, a boundary this study characterizes but does not yet resolve.
Keywords:	Overlapping Community Detection, Leader-Based Algorithms, Attributed Networks, Fuzzy Membership, Information Matrix
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	AN INTEGRATED FINBERT–TEMPORAL FUSION TRANSFORMER WITH DYNAMIC RISK-AWARE LOSS FOR STOCK FORECASTING AND TRADING DECISION SUPPORT
Author:	S. VALLI, DR S. SIVASUBRAMANIAM
Abstract:	Accurate stock price prediction remains a challenging task because of the high volatility, nonlinearities, and diverse sentiment characteristics of financial markets. Conventional forecasting approaches focus on the historical behavior of stock prices and ignore the effects of textual sentiment, financial risk, and their combined influences during the optimization process. This study presents a novel multimodal stock forecasting framework that simultaneously considers Fin BERT based financial-news representations, market and technical indicators, Temporal Fusion Transformer modelling, and a Dynamic Risk-aware Loss function that regularizes the prediction error, volatility, Conditional Value-at-Risk (CVaR), and prediction direction. This framework extends beyond accuracy-oriented stock forecasting towards enhanced risk estimation, explainability, and decision making in terms of BUY, HOLD, or SELL signals. First, we enrich the input space formed by historical stock prices with eighteen technical indicators, Fin BERT news representations, and market volatility features resulting in 794 input variables. Then, we propose a Temporal Fusion Stock Forecasting Framework (TFSF) based on a hybrid CNN-BiLSTM-TFT architecture to estimate future stock prices. We evaluate our framework on four India-based stock market datasets including CIPLA, HDFC Bank, Infosys, and ONGC after applying a rigorous preprocessing and sliding-window procedure. Our experiments demonstrate that our approach outperforms CNN, LSTM, BiLSTM, CNN-BiLSTM, TFT, and conventional TFSF models in terms of the RMSE, MAE, MAPE, and R2 evaluation metrics with the RMSE = 0.0101, MAE = 0.0079, MAPE = 1.09%, R2 = 0.991, respectively. In addition, our approach achieves higher financial performance with a Sharpe Ratio = 1.94, and Value at Risk (VaR) = 2.10%, compared to the baseline methods while providing accurate BUY, HOLD, and SELL recommendations. Finally, we provide an interpretability analysis showing the influence of each feature on the final prediction where the highest importance is assigned to Fin BERT features. Overall, the proposed approach provides an accurate, interpretable, risk-aware, and intelligent framework for financial forecasting and decision making.
Keywords:	Stock Price Forecasting, Temporal Fusion Transformer (TFT), Fin BERT-Dynamic Risk-Aware Loss, Financial Sentiment Analysis
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	ADAPTIVE TEMPORAL-SPATIAL MULTIMODAL ATTENTION NETWORK FOR EARLY BRAIN DISEASE PREDICTION USING MRI, FMRI, AND EEG
Author:	JEROMY R , Dr. JEBAMALAR TAMILSELVI J
Abstract:	Early and accurate prediction of brain diseases is essential for timely diagnosis, intervention, and effective clinical management. Neurological disorders are associated with complex changes in brain structure, functional connectivity, and electrophysiological activity, which can be captured using Magnetic Resonance Imaging (MRI), functional Magnetic Resonance Imaging (fMRI), and Electroencephalography (EEG), respectively. However, existing brain disease prediction approaches mainly depend on individual modalities, thereby limiting their ability to capture complementary disease-related information. Conventional multimodal approaches also frequently employ static feature fusion, which may not adequately adapt the contribution of structural, functional, and temporal information according to disease characteristics and progression. To address these challenges, this research proposes an Adaptive Temporal-Spatial Multimodal Attention Network (ATSMAN) for early brain disease prediction through adaptive integration of MRI, fMRI, and EEG information. Initially, the multimodal data are preprocessed to reduce noise and ensure data consistency. Subsequently, modality-specific Deep Learning (DL) architectures are employed to extract discriminative representations. A Three-Dimensional Convolutional Neural Network (3D-CNN) is utilized to learn spatial and structural features from MRI data, while a Graph Attention Network (GAT) constructs and analyzes functional connectivity representations from fMRI data. In parallel, a Bidirectional Long Short-Term Memory-Transformer (BiLSTM-Transformer) architecture captures temporal dependencies and electrophysiological patterns from EEG signals. The extracted structural, functional, and temporal representations are then integrated through an adaptive temporal-spatial attention mechanism, which dynamically assigns modality-specific relevance weights and generates a unified multimodal representation. The fused representation is subsequently processed by the disease prediction module to produce disease probabilities, predicted labels, and confidence scores. Finally, an Explainable Artificial Intelligence (XAI) module identifies important anatomical regions, functional connections, and temporal EEG patterns associated with the prediction. Thus, ATSMAN provides comprehensive multimodal integration, adaptive biomarker selection, and interpretable prediction, offering a promising framework for early brain disease prediction and precision neurological assessment.
Keywords:	Early Brain Disease Prediction, Multimodal Learning, MRI, fMRI, EEG, Adaptive Attention Network, DL, GAT, Explainable Artificial Intelligence, Neurological Disorders
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	A NOVEL SELF-ADAPTIVE GRAPH TRANSFORMER FOR EXPLAINABLE BUFFER OVERFLOW VULNERABILITY DETECTION THROUGH DYNAMIC CODE SEMANTICS AND EVOLUTIONARY OPTIMIZATION
Author:	S. THENMOZHI,mDr. PM. SHANTHI
Abstract:	The Buffer overflow vulnerabilities continue to pose significant security challenges in modern software systems, frequently resulting in memory corruption, privilege escalation, denial-of-service attacks, and remote code execution. Traditional vulnerability detection techniques primarily rely on static code analysis, handcrafted features, or conventional Graph Neural Networks (GNNs), which often struggle to capture dynamic execution semantics, long-range code dependencies, and complex structural relationships. These limitations can reduce detection accuracy, increase false-positive rates, and provide limited interpretability for security analysts. To overcome these challenges, this paper proposes a Self-Adaptive Graph Transformer Framework for Explainable Buffer Overflow Vulnerability Detection via Dynamic Semantic Graph Learning and Evolutionary Feature Optimization. The proposed framework constructs a unified semantic graph by integrating Abstract Syntax Trees (ASTs), Control Flow Graphs (CFGs), Data Flow Graphs (DFGs), Program Dependence Graphs (PDGs), and dynamic execution traces, thereby preserving both structural and runtime semantic information. A Self-Adaptive Graph Transformer (SAGT) employing multi-head self-attention learns contextual relationships among code entities while adaptively assigning attention weights according to execution semantics and vulnerability relevance. An Evolutionary Feature Optimization (EFO) module identifies informative semantic graph features and eliminates redundant information. An Explainable Artificial Intelligence (XAI) module further provides node-level and graph-level explanations, feature importance rankings, and vulnerability localization. The proposed framework is evaluated on publicly available software vulnerability datasets and benchmark repositories using Accuracy, Precision, Recall, F1-score, Matthews Correlation Coefficient (MCC), ROC-AUC, PR-AUC, and False Positive Rate (FPR). Experimental results demonstrate that the proposed framework outperforms the evaluated static analysis, GNN, Transformer-based, and deep learning approaches in vulnerability detection performance. The integration of dynamic semantic graph learning, adaptive graph attention, evolutionary feature optimization, and explainable decision-making provides a promising framework for automated buffer overflow vulnerability detection and intelligent software security applications.
Keywords:	Buffer Overflow Vulnerability Detection, Graph Transformer, Self-Adaptive Learning, Dynamic Code Semantics, Explainable Artificial Intelligence (XAI), Deep Learning, Vulnerability Classification, Intelligent Secure Software Development.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	A BECKN-ENABLED INTEROPERABLE FRAMEWORK FOR DECENTRALIZED SMART AGRICULTURE ECOSYSTEMS
Author:	MANJU SADASIVAN , ASHOK KUMAR T A
Abstract:	The agricultural sector is plagued with barriers to accessing the market, inefficiencies to supply chain, and fragmentation of data. Traditional Digital solutions catering to the agricultural sector operate in independent silos, forming a much-disconnected environment to the farmers, service providers, and consumers. This article proposes a new smart agriculture framework based on the Beckn Protocol, an open-source, interoperable specification for decentralized digital commerce. The approach adopted is mapping of key agriculture processes such as input procurement, crop monitoring, and produce sale to the corresponding discovery, order and fulfillment flows in the Beckn protocol. The primary goal is to establish an open agricultural network that enables seamless service discovery, ordering, and fulfillment across heterogeneous providers through a single application interface. The study hypothesizes that a Beckn-enabled framework will significantly reduce workflow completion time, increase service discovery rates, and enhance farmer autonomy compared to centralized siloed platforms. The framework is tested using a discrete event simulation parameterized using ONDC benchmarks, a farmer survey conducted in Karnataka, India and public eNAM datasets. The results show a 61.3% reduction in time for completing the workflow, a successful service discovery rate of 94.2% and a farmer autonomy rate of 41.2%. The framework also delivers a 6.2x benefit in price transparency and scales SLA compliant with a latency of p95. In addition, it cuts the transaction costs by 70-90%. The proposed framework is different from blockchain-based solutions and ONDC as it provides agriculture-specific schemas, discovery in a decentralized manner, and low-cost transactions in a consistent eco-system. This study contributes the first domain-specific adaptation of the Beckn Protocol for agriculture, novel schema extensions, and empirical evaluation of decentralized agricultural service ecosystems. Future research will explore the integration of reinforcement learning, voice-assisted interfaces, and blockchain-based traceability to further enhance the framework's capabilities. Overall, the Beckn-enabled framework provides a sound empirical approach to open, interoperable and farmer-centric digital agriculture
Keywords:	Agri-tech, Beckn Protocol, Decentralization, Digital Ecosystems, Interoperability, Open Networks, Smart Agriculture Component.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	QUALITATIVE ANALYSIS OF PHARMACEUTICAL COLD CHAIN FAILURES AND DIGITAL MONITORING TECHNOLOGIES: REGULATORY IMPLICATIONS FOR PATIENT SAFETY IN MOROCCAN HEALTHCARE FACILITIES
Author:	JAD EZZAHER, RIHAB EZZAHER, YASSIN SELOUAN, ABDELLAH MARGHICH
Abstract:	The integrity of the pharmaceutical cold chain is a determining condition for the safety, efficacy and legal conformity of thermolabile medicinal products such as vaccines, insulins, biologics and blood-derived preparations. Empirical evidence gathered in low- and middle-income countries indicates that cold chain breaks are frequent, largely undetectable at the point of care, and closely associated with weak regulatory enforcement, resource-constrained infrastructure and organisational cultures that discourage incident reporting. Digital monitoring technologies - continuous data loggers, vaccine vial monitors and integrated traceability platforms - increasingly condition the practical enforceability of cold chain regulation, since they convert compliance from a matter of retrospective professional judgement into one of contemporaneous, auditable record. Building on a conceptual framework defining the pharmaceutical cold chain, cold chain disruption, patient safety, good distribution practices, quality risk management, digital monitoring technology and regulatory compliance, this article first reviews eleven empirical studies conducted in Ethiopia, a thirteen-country sample of low- and middle-income countries, Nigeria, Poland, India, the United States, Portugal, Zambia, Senegal and China, and analyses Morocco's evolving pharmaceutical regulatory architecture, marked by the transition from the 2006 Code du Médicament et de la Pharmacie (Law No. 17-04) toward the newly created Moroccan Medicines and Health Products Agency (AMMPS) and the pending reform bill (Law No. 27.26) that seeks alignment with the World Health Organization's Maturity Level 3 benchmark. It then reports a qualitative investigation based on semi-structured interviews conducted with eight hospital pharmacists practising in the Fez-Meknes region, organised around four analytical axes covering the identification of cold chain disruptions, their organisational and infrastructural determinants, incident management and reporting culture, and the perceived adequacy of the regulatory framework. The findings suggest that cold chain failures in Moroccan public hospitals are perceived as multidimensional and systemic rather than attributable to individual negligence, shaped by ageing equipment, staffing constraints, uneven adoption of digital temperature-monitoring devices, uneven reporting practices and a persistent gap in operational guidance for managing disruptions once they occur - a gap that the anticipated Law No. 27.26 could help to close. The article concludes by situating these findings within the comparative literature and by outlining directions for further regional and quantitative research.
Keywords:	Pharmaceutical Cold Chain; Cold Chain Disruption; Digital Monitoring Technology; Patient Safety; Good Distribution Practices; Quality Risk Management; Hospital Pharmacists; Pharmaceutical Regulation; Morocco.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	A TRUSTWORTHY MULTIMODAL FOUNDATION FRAMEWORK WITH ADAPTIVE CROSS-MODAL ATTENTION FUSION FOR OPEN-WORLD TOMATO DISEASE DIAGNOSIS IN REAL AGRICULTURAL ENVIRONMENTS
Author:	RAJI NETTATH , DR. VIJAYABHANU RAJAGOPAL
Abstract:	Diseases are important factors which threaten the productivity of tomato crops worldwide and result in heavy yield losses, monetary losses to other crops and pose serious challenges. Despite the recent progress of automated diagnosis of plant diseases using deep learning methods using only RGB images, most of the current deep learning approaches are only based on RGB images, require closed-set environments, do not account for environmental variability, and are unable to account for prediction reliability in real-world field environments or to deal with unknown diseases. Considering the above issues, the present paper introduces a novel Trustworthy Multimodal Foundation Framework (TMFF) for open world tomato disease diagnosis for precision agriculture. The proposed framework includes a pre-trained vision foundation model and various heterogeneous environmental factors like weather data, soil characteristics, crop growth stage, and Internet of Things (IoT) sensor data to learn powerful multimodal representations of diseases. Propose an adaptive cross-modal attention fusion module (ACMAF) that can adaptively learn the complementary interactions between the visual and environmental features through bidirectional attention and adaptive modality weighting. Also, an Open-World Recognition Module (OWRM) is capable of reliably recognizing known and unseen disease categories based on prototype-based representation learning and adaptive threshold estimation. In order to improve trust in decision-making, a Trustworthiness Assessment Module (TAM) will have Bayesian uncertainty estimation, calibrated confidence and explainable artificial intelligence-based prediction. Experimental evaluation shows that TMFF's classification achieves 99.18%, F1-score achieves 99.09%, macro-AUC achieves 0.9987 and unknown disease detection achieves 97.88%, which is better than state-of-the-art convolutional neural networks, vision transformers and foundation models while allowing real-time inference. According to this proposal, an integrated reliable and accurate tomato disease diagnosis solution can be developed. This will have great potential applications for precision agriculture and next-generation artificial intelligence-enabled crop health monitoring.
Keywords:	Tomato Disease Diagnosis; Foundation Models; Multimodal Learning; Adaptive Cross-Modal Attention Fusion; Open-World Recognition; Trustworthy Artificial Intelligence; Vision Transformers; Precision Agriculture; Explainable Artificial Intelligence; Edge Intelligence.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	ACH-CIR: ATTENTION-GUIDED HYBRID CNN–ANN FRAMEWORK FOR SECURE CANCELABLE IRIS RECOGNITION IN CLOUD COMPUTING ENVIRONMENTS
Author:	P. SHABANA , Dr R BALA KRISHNA
Abstract:	Iris biometric authentication systems employ the exclusive texture patterns of the iris to identify an individual; they can be very stable over time and are well suited for secure identity verification applications. Protecting stored iris templates is difficult, however, because of privacy concerns, irreversible exposure to iris biometrics and diminished authentication confidence if the templates are compromised. For iris recognition, researchers have developed new techniques based on deep learning to achieve better performance in feature extraction and classification. Frequently, however, these methods have weak template security, lack of feature optimization, and lack of scalability for authentication in cloud-based environments. Thus, researchers proposed the ACH-CIR: Attention-Guided Hybrid CNN–ANN Framework for Secure Cancelable Iris Recognition in Cloud Computing Environments. The pre-processing of iris images is done by segmentation, normalization and enhancement techniques in ACH-CIR model which is based on USIT framework. It then transforms the data using secure cancelable transformation through BLAKE3 hashing, generation of user-specific seed-matrix, and inverse feature transformation for privacy preservation. Convolutional Neural Networks (CNN) are used to extract features and an attention mechanism is further used to optimize the features. The hybrid deep learning integration consists of extracting the features using CNN and authentication using ANNs for iris authentication. Such metrics as accuracy, precision, recall, GAR, FAR, FRR and EER are used to measure performance. The proposed Attention-guided CNN–ANN model is able to provide better recognition accuracy and protection for template in standard CASIA iris datasets. The method could lead to a substantial enhancement of secure biometric authentication, as it allows for accurate, revocable and scalable iris recognition in cloud computing environments.
Keywords:	Cancelable Iris Recognition (CIR) System, Alongside The Use Of CNN, ANN, Attention Mechanism, BLAKE3 Hashing, Cloud Authentication, And Biometric Security.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	ARCHITECTURE OF INTELLECTUAL IOT-SYSTEM OF DATA COLLECTION FROM WEARABLE DEVIСES FOR PREVENTIVE HEALTH STATUS MONITORING
Author:	ANDRIY PIDDUBNYY, ANDRIY STRAZHNIKOV, OLEKSANDR PRONKIN, HALYNA KONDRATSKA, OLEKSANDR ROTAR
Abstract:	The purpose of the study is to provide analytical substantiation and comparative analysis of architectural solutions for biometric data collection from wearable devices. The work analyses the hypothesis according to which the level of integration of peripheral calculations (Edge) and the type of adaptive filtering algorithms are decisive factors in the accuracy of biometric parameters. Deconstruction of architectural models was conducted within the context of Industry 5.0, which involves methods of decentralised signal processing and hybrid data transmission protocols on the example of a case-study of Garmin, Our, and Apple Watch systems. The evaluation is based on the comparative analysis of root mean square error (RMSE) and latency in scenarios of physical activity and rest. The analysis results demonstrate that architectures with Edge-processing priority ensure deviation reduction of 35-58% and stabilisation at the level of 100ms compared to purely cloud solutions. The data confirm the high effectiveness of algorithms of intellectual adaptive filtering at levelling motion artefacts directly at the peripheral gateway. The conclusions systematise the advantages of analytical load delegation to the peripheral level to ensure scalability and sustainability of preventive healthcare systems. Further study perspectives lie in exploring methods of lightweight machine learning for monitoring personalisation in real time. The use of decentralised models enables the extension of the possibility of remote rehabilitation under the condition of preservation of high reliability of biometric parameters.
Keywords:	Edge Computing, Smart Rings, Physical Rehabilitation, Physiological Signal Processing, Photoplethysmography (PPG), Artifact Removal, Industry 5.0.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	ADAPTIVE FROST-FILTERED DEEP FEATURE FUSION WITH METAHEURISTIC-OPTIMIZED WGAN-AUTOENCODER FOR IOT-ASSISTED SKIN CANCER DETECTION
Author:	MUTHAMIZHAN M, B SATHYASRI
Abstract:	Internet of Things (IoT) aided skin cancer detection incorporates many linked sensors and devices to support the main study and observation of skin modalities. Skin cancer is measured as one of the hazardous types of cancer and there is a severe upsurge in the number of deaths owing to a low level of awareness of the signs and their prevention. Therefore, early identification at a primary phase is essential so that one can avoid the increase of cancer. Recently, IoT-based skin cancer detection using deep learning (DL) was employed for improving the prompt analysis and observation of skin cancer. This paper presents a Hybrid Deep Learning and Feature Extraction Technique for Enhanced Skin Cancer Diagnosis through Medical Imaging (HDLFET-ESCDMI) model. The aim of the paper is to develop an IoT-assisted medical image analysis framework for accurate skin cancer diagnosis using advanced techniques. Initially, the image pre-processing stage employs an extended-tuned adaptive frost filtering (Ext-AFF) technique for eliminating the noise to enhance image quality. Besides, the extraction of the feature process has been executed by the ConvNeXt-X model to recognize and isolate the most significant information from raw data. For the classification procedure, the presented HDLFET-ESCDMI model implements a hybrid Wasserstein generative adversarial network and autoencoder (WGAN-AE) technique. At last, the red-billed blue magpie algorithm (RBMO)-based parameter fine-tuning procedure is implemented to enhance the classification outcomes of the WGAN-AE system. A comprehensive experiment was implemented to verify the performance of the HDLFET-ESCDMI. The performance results designated that the HDLFET-ESCDMI underscored advancement over other recent techniques.
Keywords:	Hybrid Deep Learning, Feature Extraction, Skin Cancer Diagnosis, Medical Imaging, Internet of Things, Hyperparameter Tuning
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	A PROPOSED FRAMEWORK FOR TECHNOLOGY INTEGRATION IN PUBLIC EDUCATION: AN EMPIRICAL STUDY OF TEACHERS’ CHALLENGES AND SOLUTIONS
Author:	HANEH RHEL, REEM ALI SALEM ABDALLA, ASMA ALI MOSA ELTHARIF, AMIRA MUSTAFA
Abstract:	Integrating technology into the educational process has become essential for improving teaching and learning practices. However, its effectiveness depends to a large extent on the ability of teachers to utilize digital tools within classrooms. Despite the efforts made, many educational institutions still face difficulties that limit the optimal use of technology. This study examined the most significant challenges associated with integrating technology into public schools in Benghazi and explored whether these challenges varied according to gender, specialization, academic qualification, and years of teaching experience. A descriptive-analytical approach was employed. Data were collected through a questionnaire distributed to 171 teachers in preparatory schools during the 2023-2024 academic year. The results showed that the challenges related to technology integration were widespread (M = 3.90), with challenges identified at both the teacher and school level. No statistically significant differences were found according to gender, specialization, educational qualification or years of teaching experience. This suggests that these obstacles are structural rather than individual in nature. Based on these results, a data-driven framework was proposed, drawing on the Educational Technology Content Knowledge Model (TPACK) and systems theory. This framework provides a structured and sustainable approach for improving the integration of technology in education, through practical guidelines relating to professional development, infrastructure development, curriculum adaptation, and continuous assessment. Its role is not limited to improving the efficiency of technology use within classrooms, but also provide a basis for further testing in other educational contexts facing similar systemic challenges.
Keywords:	Technology Integration, TPACK Model, Systems Theory, Educational Challenges, Benghazi Schools.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	AGRIRL: AN ATTENTIVE DEEP LEARNING AND REINFORCEMENT LEARNING FRAMEWORK FOR SIMULATOR-BASED ADAPTIVE FERTILIZER RECOMMENDATION
Author:	PIDUGU NAGENDRA, DR. CH. VENKATA RAMANA REDDY, DR. K. PRADEEP REDDY
Abstract:	Precision fertilizer management requires both accurate diagnosis of the current soil-crop state and sequential decisions that adapt to changing nutrient status, weather, and crop demand. Most public-data studies formulate fertilizer recommendation as a one-shot classification problem, whereas reinforcement-learning studies typically optimize actions inside a crop simulator. This study proposes AgriRL, a two-layer framework that deliberately separates these two forms of evidence: (i) an attentive tabular deep classifier for fertilizer-class prediction and (ii) a finite-horizon deep Q-network (DQN) for stage-wise fertilizer type-and-dose selection in a transparent surrogate crop-response environment. The supervised layer was evaluated on a frozen 10,000-record public dataset with five harmonized classes using a stratified 70/15/15 train/validation/test split, training-only imputation and scaling, fold-restricted SMOTE, and five-fold hyperparameter tuning on the training partition. The sequential layer was evaluated independently on 50 paired six-stage simulated seasons in which all compared policies received identical initial states and weather realizations. AgriRL achieved 98.7% held-out classification accuracy and 98.5% macro-F1. In the retained simulator, mean seasonal yield increased from 5.87 ± 0.24 t/ha under the traditional schedule to 6.58 ± 0.20 t/ha under AgriRL, corresponding to a paired mean difference of 0.71 t/ha (12.1%). Ablation results indicate that attentive representation and class balancing primarily improve classification, whereas the DQN primarily changes sequential simulator outcomes; reward-term ablation further shows the cost-risk trade-off of yield-only optimization. Because no field trial or independently calibrated crop model was available, all productivity, cost, nutrient-use-efficiency, and environmental-risk outcomes are reported strictly as simulator-derived evidence. The principal scientific contribution is therefore a reproducibly specified and statistically auditable bridge between interpretable fertilizer classification and adaptive sequential decision-making, with explicit separation of observed-data performance from simulated agronomic outcomes.
Keywords:	Adaptive Fertilizer Management, Attentive Tabular Learning, Deep Q-Network, Precision Agriculture, Simulator-Based Evaluation
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	STEP-FORMER: RELIABILITY-AWARE SPATIO-TEMPORAL POSE MODELING FOR LIGHTWEIGHT HUMAN ACTION RECOGNITION UNDER IMPERFECT POSE ESTIMATES
Author:	B. PANDU RANGA RAJU, DR. CH. VENKATA RAMANA REDDY, DR. K. PRADEEP REDDY
Abstract:	Human action recognition using pose decreases the importance of appearance cues and allows for privacy-aware recognition; however, recognition accuracy may decrease due to missing, jittered, or inconsistently detected two-dimensional poses. Recent graph-, transformer-, temporal-pooling-, hypergraph-, and language-assisted skeleton approaches are commonly evaluated with different skeleton dimensionalities, modalities, datasets, or preprocessing pipelines; consequently, their published scores are not directly comparable with a controlled 2D OpenPose setting. In this work, we propose STeP-Former, a lightweight pose-only recognizer that uses Spatio-Temporal Pose Abstraction Layer (ST-PAL), label-free Reliability-Aware Temporal Fusion (RTF), and shallow temporal attention-based encoder. ST-PAL maps normalized joint coordinates and their first order derivatives to motion tokens. RTF computes the reliability of individual frames from the consistency of local tokens and penalizes unreliable frames before global temporal modeling. The evaluation is carried out following the usual disjoint protocol for KTH dataset and the official UCF101 Split 1. Four baseline models that use poses are reproduced on the same cached OpenPose BODY-25 sequences with three fixed seeds, while RGB-based or protocol non-compatible recent approaches are provided separately for reference purposes as literature only. STeP-Former achieves 95.6 ± 0.28% accuracy on KTH and 90.2 ± 0.41% accuracy with 0.897 ± 0.007 macro-F1 score on UCF101. The recognizer has 2.1 M parameters and takes 9 ms to process 32 frames on a clip according to the mentioned recognizer-only timing protocol. Under the two controlled empirical stress conditions, STeP-Former shows smaller accuracy degradation than the aggregate pose baselines, while the simulation-only results are retained as sensitivity analyses rather than benchmark evidence.
Keywords:	Human Action Recognition, Two-Dimensional Pose Sequences, Motion Tokens, Reliability-Aware Temporal Fusion, Temporal Attention, Recognizer Efficiency
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	ECO PESTNET: AN ECO-CONSCIOUS HYBRID AI FRAMEWORK FOR PRECISION PEST DETECTION AND SUSTAINABLE PESTICIDE INTERVENTION PLANNING
Author:	SILIVERU ASHOK KUMAR, DR. CH. VENKATA RAMANA REDDY, DR. K. PRADEEP REDDY
Abstract:	Accurate pest classification alone is insufficient for sustainable pest intervention because an operational decision system must connect visual evidence with environmental context, spatial infestation patterns, ecological risk, dosage, and route selection. This study proposes EcoPestNet, a hybrid framework that combines UAV imagery, IoT environmental sensing, attention-guided convolutional features, graph-convolutional spatial reasoning, an ecological-risk score, adaptive dosage, deep Q-learning for mission planning, and visual/sensor explanations. The principal methodological contribution is an explicitly source-separated perception-to-intervention formulation: public-image data are used to evaluate perception, while independent simulated missions are used to evaluate intervention behavior, so diagnostic and action evidence are not conflated. On the PlantVillage-Corn evaluation, EcoPestNet achieved 93.8 ± 1.1% accuracy, 94.1% precision, 93.2% recall, 93.6% F1-score, 0.968 ROC-AUC, and 0.957 PR-AUC. Across 500 held-out 20 × 20 simulator missions, the full framework reduced the normalized pesticide-volume index by 15.4% and redundant spray overlap by 18.7% relative to the fixed-dose baseline. Ablation results showed a 2.1 percentage-point accuracy advantage over the no-GCN variant and substantial deterioration of intervention savings when the DQN planner or ecological penalty was removed. These findings establish the study's contribution as an auditable integration of spatially informed perception and eco-aware intervention planning, while the intervention results remain simulation-based and are not claims of physical field-spraying outcomes.
Keywords:	Precision Agriculture, Pest Detection, UAV Imaging, Graph Convolutional Network, Deep Reinforcement Learning
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	HYBRID DEEP LEARNING AND ENSEMBLE STRATEGY FOR ALZHEIMER’S DETECTION USING YOLOV12 AND RFXGBOOST
Author:	Dr. MAGANTI VENKATESH, Dr SHAIK JANBHASHA, E. RAVEENDRA REDDY, Dr P DAYAKER, KODIPAKA VENKATESHWAR RAO, VEERAMOHANA RAO REDDY, MADHAN KUMAR JETTY, Dr KODIPAKA RAJESHWAR RAO
Abstract:	Alzheimer disease (AD) is an irreversible, progressive neurodegenerative disorder that drastically impairs memory, thinking and behavioral functions. Early AD diagnosis is important in ensuring clinical intervention is effective and a patient has a good quality of life. Nevertheless, manual neuroimaging evaluation and traditional diagnostic methodologies are usually time consuming, subjective and are subject to human error. In addition, most current machine learning and deep learning models fail to be able to represent high-dimensional biomedical data, which can reduce the accuracy of diagnostic results. To overcome these issues, we suggest an efficient, entirely automated diagnostic model, which incorporates deep-learning with ensemble-learning to make an accurate classification of Alzheimer disease and its progression. The given system has three major steps namely: data acquisition, feature extraction, and classification. Neuroimaging data is pre-processed to be consistent and of quality. YOLOv12 is then used as a deep feature extractor backbone, where automatic learning of discriminative spatial and structural features that are of interest to pathology of Alzheimer are learned. Such extracted features are then categorized with the help of a hybrid (RF+XGboost) of Random Forest (RF) and XGBoost (RF+XGboost). Such combination uses the advantages of the RF in terms of variability of features and XGBoost in terms of the ability to capture the complex nonlinear relationships and increases the robustness and predictive performance. The experimental findings show that the YOLOv12 + RF+XGboost model is much better than the traditional deep learning models. The proposed system achieves an accuracy of 98.72%, precision of 98.55%, recall of 98.41%, and F1-score of 98.48%. These results prove the practicality of the YOLOv12+RFXGBoost hybrid model that allows making effective early diagnosis and progression measurements of Alzheimer disease and provides a promising track toward better clinical decision-making and patient care when compared with ResNet-101(92%) and VGG19(88.72%).
Keywords:	Hybrid deep learning, Ensemble learning, Neuroimaging-based progression, Alzheimer’s disease. ResNet-101, VGG19, RF+XGBoost.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	SWIPE RIGHT, DRIFT AWAY: A SYSTEMATIC REVIEW OF SOCIAL MEDIA FACILITATED THIRD-PARTY INTERFERENCE IN MARRIAGE AND ITS PSYCHOLOGICAL CONSEQUENCES FOR COUPLES
Author:	AINI AZEQA MAROF, HASLINDA ABDULLAH, HANINA H. HAMSAN
Abstract:	Digital communication technologies have fundamentally transformed the landscape of third-party interference in marriage, creating new pathways through which extramarital emotional and romantic involvement emerges, escalates, and produces marital distress. Despite a growing body of research on social media use and relationship outcomes, no systematic review has comprehensively synthesised the evidence on social media facilitated third-party interference as a distinct and coherent phenomenon, examined the specific mechanisms through which digital environments enable and amplify relational violations, or integrated this evidence with established frameworks of betrayal trauma and marital distress. This systematic review addresses these gaps by synthesising empirical research published between 2009 and 2024 on the association between social media use and third-party interference in marriage, the psychological consequences for affected couples, and the mechanisms through which digital contexts facilitate relational boundary erosion. Following PRISMA 2020 guidelines, five academic databases were systematically searched, namely PsycINFO (n = 198), Scopus (n = 224), Web of Science (n = 167), PubMed (n = 89), and Communication Abstracts (n = 134), yielding a total of 812 records and ultimately 16 eligible empirical studies after deduplication and screening. Findings reveal that social media facilitated third-party interference operates through four primary mechanisms: parasocial intimacy escalation, digital disinhibition, privacy architecture exploitation, and parallel reality construction. Psychological consequences for affected partners include heightened relational jealousy and surveillance behaviour, betrayal trauma symptoms equivalent in severity to offline infidelity, accelerated trust erosion, and partner-directed hypervigilance. The review makes four theoretical contributions: the Digital Intimacy Gradient Model, which maps the escalation pathway from casual online contact to emotionally intimate third-party involvement across five progressive stages; the concept of ambient digital infidelity, describing the continuum of digitally mediated relational boundary violations that precede formal affair disclosure; the Dual Screen Relationship Model, which conceptualises marriage in the digital age as a relationship system simultaneously maintained in physical and digital spaces; and a Digital Boundary Literacy Framework as a prevention-oriented construct with direct implications for premarital education, couple therapy, and digital wellness policy. Implications for therapeutic practice, preventive intervention, and family policy in high-connectivity societies are discussed.
Keywords:	Digital Infidelity, Online Affair, Marital Distress, Betrayal Trauma, Internet Infidelity
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	NARX-BASED VIRTUAL IMU SENSOR FOR FAULT-TOLERANT ATTITUDE ESTIMATION AND CONTROL OF QUADROTOR UAVS
Author:	BUI THANH TUNG, TRAN THI VAN, NGUYEN VAN THANG, KHOA NGUYEN DANG
Abstract:	Accurate attitude estimation is a fundamental requirement for achieving stable flight and reliable control of quadrotor unmanned aerial vehicles (UAVs). However, failures or abnormal measurements of the Inertial Measurement Unit (IMU) may interrupt attitude feedback and severely degrade flight performance. This paper proposes a Virtual IMU Sensor (VIMUS) based on a Nonlinear AutoRegressive network with eXogenous inputs (NARX) to estimate roll and pitch angles from control inputs and quadrotor dynamics. To evaluate its effectiveness, a Multilayer Perceptron (MLP)-based virtual sensor is implemented as a benchmark under identical training conditions. An IMU fault detection and switching mechanism is further integrated to automatically replace the original IMU with the proposed VIMUS when sensor faults occur. Simulation results show that the NARX-based VIMUS outperforms the MLP model, reducing the root mean square error (RMSE) by 62.3% and 64.6% for roll and pitch estimation, respectively, while achieving coefficients of determination (R²) of 0.986 and 0.991. The proposed framework enables continuous attitude feedback and provides an accurate and robust fault-tolerant solution for quadrotor attitude estimation and control.
Keywords:	NARX Neural Network, Multilayer Perceptron, IMU Fault Detection, Quadrotor UAV, Attitude Estimation
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	HYBRID QUANTUM-CLASSICAL OPTIMIZATION MODEL FOR SMART GRID LOAD FORECASTING AND SECURE ENERGY TRADING
Author:	N VENKATA SAILAJA, DR. NAGARATNA P HEGDE2, DR. SIREESHA VIKKURTY, VIJAYA CHANDRA JADALA, M. SHANMUGA SUNDARI, KBKS DURGA
Abstract:	The increasing integration of renewable energy sources and decentralized energy trading has significantly increased the complexity of modern smart grid management. This study proposes a hybrid quantum–classical optimization framework for smart grid load forecasting and secure energy trading. A two-layer Long Short-Term Memory (LSTM) network is used for load prediction, achieving an RMSE of 2.89 MW and MAPE of 3.71%, improving forecasting accuracy by approximately 9% compared to classical models. The economic dispatch problem is formulated as a Quadratic Unconstrained Binary Optimization (QUBO) model and solved using the Quantum Approximate Optimization Algorithm (QAOA). Experimental evaluation on the IEEE 33-bus system reduced total dispatch cost to $12,310/day, achieving a 5.2% cost reduction compared to PSO. The framework also ensures secure peer-to-peer energy trading using blockchain with post-quantum cryptography. The results demonstrate improved scalability, optimization efficiency, and security for next-generation smart grid systems.
Keywords:	Hybrid quantum–classical optimization, Post-quantum cryptography, Quadratic Unconstrained Binary Optimization (QUBO), Quantum Approximate Optimization Algorithm (QAOA), Secure energy trading, Smart grid load forecasting.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	AN INTELLIGENT MACHINE LEARNING FRAMEWORK FOR ANOMALY AND SECURITY ATTACK DETECTION TO ENABLE SECURE AUTHENTICATION
Author:	KALYAN KUMAR DASARI, Dr.K. SAHADEVAIAH
Abstract:	The authentication process entails transfer of sensitive data between the device or account of the user and the authentication server in a bid to identify an identity of the requester. The authentication process needs to be secured against security assaults. Authentication techniques are numerous depending on the requirements of the system. However, any authentication method cannot be fully secure. This study was aimed at identifying the security threats and abnormalities by employing the device and user context in the authentication process. Context has also been used to examine the Denial-of-service as DoS, attacks like brute-force and distributed denial-of-service (DDoS) attacks. Extensive simulations happening on the standard dataset is CIC-IDS2018 were conducted with python programming. The accuracy, recall, f-score precision and also built time of each of four ML classifiers that are decision tree classifier model, Random Forest classifier, k-nearest neighbour classifier and SVM model were computed on the various combinations of splits of data and various features splits. Approximately the tests give the f-scores, recalls, precisions, and accuracy of over 98.0 of brute-force and DoS/DDoS attacks. Discoveries on the tests security and safety analysis and threat modelling protest the projected validation and authentication strategy can be used to raise the level of security of a secure system.
Keywords:	Sensitive Information, Authentication, DDoS attacks, Machine learning classifier, Brute force, Random Forest.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	DEEP SEMANTIC ONTOLOGY WITH CASE-BASED REASONING FOR LEGAL DOCUMENT AUTOMATION AND HYBRID RETRIEVAL-SUMMARIZATION
Author:	NAIMOONISA BEGUM, G. REKHA
Abstract:	As more of the judicial documents are digitized and the information becomes more complex, it has become more difficult to provide accurate information retrieval and summarization in legal documents. While current keyword-based retrieval systems often miss out on semantically similar, but lexically distinct, precedents, generic neural summarization models may struggle to capture legal reasoning, statutory context, and long-range context. The restrictions are especially relevant in the Indian context where judgment writing is lengthy, structurally varied, precedents dependent and so is the influence of the specific terminology in these judgments. This paper presents the Legal Aware BigBird Longformer Encoder–Decoder (LawBird-LED), a transformer-based hybrid model designed for intelligent legal retrieval and automated legal summarization to tackle these challenges. The proposed system involves using semantic ontology-based search to retrieve legally relevant information and case-based reasoning (CBR) retrieval to retrieve the precedent aware contextual knowledge for improving legal understanding. Furthermore, the structure applies both sparse attention (BigBird) and contextual attention (Longformer) for long legal documents and preserves the global relationships across the document and the local semantic relationships within the document. The summarization approach is presented as a multi-layer summarized approach to gradually improve the summarization process, and a set of interpretability techniques for the retrieval and summarization decisions is presented for transparency. Experimental comparison with the legal summarization and transformer-based benchmark models shows the proposed framework is able to generate legal summaries containing meaning and content well, and with good context capture. The proposed LawBird-LED achieved a ROUGE-F1 score of 99.0%, BLEU score of 98.8%, Legal-Sim score of 99.1%, METEOR score of 98.9% and Coverage score of 99.2%, which showed good lexical alignment, semantic consistency, and the retention of legal concepts. An effective and interpretable legal information retrieval and automatic legal text summarization system based on semantic retrieval, precedent-guided reasoning and hybrid contextual attention. The results reveal that augmenting symbolic legal knowledge with precedent-based reasoning and long-context neural representation can facilitate the semantic retrieval and legal-summary fidelity. The study therefore shows how LawBird-LED can be an interpretable decision support system for access to legal information, as well as pointing out issues of cross-jurisdictional generalization, computational efficiency and expert validation that need to be further explored.
Keywords:	Smart Legal Retrieval, Semantic Ontology, Case-Based Reasoning, Indian Jurisprudence, Automated Legal Summarization, and Legal Information Retrieval.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text

Title:	IOT-BASED AI DRIVEN MODEL FOR PREDICTION OF CARDIOVASCULAR DISEASE IN TYPE 2 DIABETIC PATIENTS USING HYBRID 1DCNN-BILSTM ARCHITECTURE
Author:	AKBERSHA K E, V. PARTHASARATHY
Abstract:	Cardiovascular Disease(CVD) is the primary cause of morbidity and mortality in Type2 Diabetes globally due to interactions between vascular and metabolic mechanisms. In these patients, disease progression is more complex, making early prediction challenging. This research implements an Integrated 1DCNN-BiLSTM model for accurate cardiac disease prediction in diabetic individuals.It comprises a 1-dimensional convolutional neural network to detect the spatial correlations of the data combine with Bidirectional Long Short-Term Memory (BiLSTM) that identify the forward and backward dependencies for an efficient analysis of sequential clinical data. The proposed model avoids the limitations of the Framingham Risk Score, which relies on manual computation and a limited set of features. The primary aim is to improve predictive accuracy while offering an economical and flexible solution for early detection, especially in resource-limited settings. Firstly,the model separate the diabetic data from Cleveland–Hungarian cardiac data sets and perform with an accuracy of 96 % for the prediction of cardiac diseases in diabetic patients. Additionally ,the model perform well over traditional methods: Support Vector Machine, Logistic Regression, Decision Tree, Naive Bayes, Random Forest and conventional 1DCNN model .The Novelty lies in transforming tabular clinical data into a sequential format to exploit temporal dependencies using BiLSTM, an approach not commonly used in traditional CVD prediction methods. Furthermore , the integrated Recursive Feature Elimination (RFE) and Principal Component Analysis (PCA) during pre-processing enhance its prediction accuracy and efficiency . Another significant contribution is that the model is applied on the NVIDIA Jetson Nano, enabling low-cost, edge-based inference for real-time IoT-enabled healthcare applications. These outcomes demonstrate that the proposed 1DCNN–BiLSTM approach is a reliable, interpretable, and practical tool for the early detection of cardiovascular risk in diabetic patients.
Keywords:	Cardiovascular Disease, Diabetes, RFE, 1DCNN, BiLSTM, ADAM.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2026 -- Vol. 104. No. 16-- 2026
Full Text