		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
February 2026 | Vol. 104 No.4

Title:	EXPLORING DIGITAL WALLET CONTINUANCE: INSTRUMENT DEVELOPMENT BASED ON TASK-TECHNOLOGY FIT THEORY AND PRIVACY CALCULUS THEORY
Author:	ADIBAH AHMAD, RAHAYU AHMAD, SYAHIDA HASSAN
Abstract:	With the growing adoption of digital wallets, increasing attention has been directed toward the challenges associated with their continued use, particularly regarding security and user privacy. This study extends prior research by integrating concerns related to technology fit and privacy through the lens of Task-Technology Fit (TTF) Theory and Privacy Calculus Theory, which serve as the study’s theoretical foundations. To address emerging factors influencing the sustained use of digital wallets, new constructs have been proposed. The measurement instruments employed in this study consist of both adapted items from existing literature and newly developed items tailored to the proposed constructs. The development process included expert evaluations to ensure content validity, involving five Information Systems scholars who assessed the relevance and clarity of the items. A pilot test was conducted involving 38 university students who participated through an online survey. Validity and reliability testing were conducted using SPSS software. Results from validity and reliability analyses demonstrate robust measurement properties, indicating that the items effectively capture the intended constructs. A subsequent pilot test further confirmed the appropriateness of the instruments for use in future empirical research.
Keywords:	Cashless, E-Wallet, TTF, PCT, Continuous Use
DOI:	https://doi.org/10.5281/zenodo.18822662
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	DUAL-MODE, RATE-AWARE SWEAT SENSING WITH UNCERTAINTY-INFORMED ANALYTICS FOR HYPERHIDROSIS SCREENING AND MONITORING
Author:	PADMA BELLAPUKONDA, DR. RAGHVENDRA KUMAR, DR. R N V JAGAN MOHAN
Abstract:	Wearable sweat sensing enables non-invasive screening and monitoring, but concentration-only readouts are confounded by secretion dynamics such as instantaneous rate and accumulated volume and by evaporation. These limitations are especially problematic in clinically important low-sweat regimes, including mild or treatment-modulated states, observed in hyperhidrosis. We present a dual-mode, rate-aware platform that fuses resistive wetting and contact conductance with a capacitive absorbent-dielectric channel to infer local sweat volume and rate, supported by a skin-interfaced microfluidic layer for chrono-sampling and evaporation compensation. On this hardware, we introduce RAISE, Rate-Aware Inference with Sensor Ensembles, an uncertainty-informed pipeline that derives rate-normalized features, applies probability calibration using slope and intercept with expected calibration error and Brier score, and quantifies clinical utility with decision-curve analysis, while reporting uncertainty using bootstrap confidence intervals. Using patient-level splits and an external temporally held-out validation cohort, our approach improved discrimination and reliability over a concentration-only baseline, with delta AUC of 0.06 and external AUC of 0.92 with 95 percent confidence interval from 0.88 to 0.95. It reduced expected calibration error to 0.031 and Brier score to 0.121, and yielded higher net benefit across clinically relevant thresholds, with maximum delta of 0.06 and up to 7 avoided interventions per 100 at probability threshold 0.10. Sensor characterization achieved mean absolute percentage error below 10 percent versus gravimetry with R squared up to 0.992 across four infusion rates. By coupling dual-mode sensing with rate-aware calibration and decision-focused analytics, the system delivers clear, clinically interpretable gains for dependable hyperhidrosis screening and monitoring.
Keywords:	Sweat Sensing; Hyperhidrosis; Microfluidics; Dual-Mode Capacitive–Resistive Sensors; Probability Calibration; Decision-Curve Analysis.
DOI:	https://doi.org/10.5281/zenodo.18822683
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	THE TIME-ORIENTED LEARNER: A FRAMEWORK FOR DESIGNING PERSONALIZED MOBILE INTERFACES TO ENHANCE TECHNOLOGY ACCEPTANCE AND SELF-EFFICACY, AND REDUCE COGNITIVE LOAD
Author:	ZHAO XIN, SITI NAZLEEN ABDUL RABU
Abstract:	Mobile learning (m-learning) design often relies on a one-size-fits-all, universal approach. This approach may create cognitive friction for some learners. This occurs when their inherent time orientation, a stable trait governing how individuals manage tasks and time, is misaligned with the interface's structure. This study introduces and evaluates Time-Oriented Personalized Interfaces (TOPI), a novel approach that aligns interface design with learners' monochronic, polychronic, or neutral dispositions. We investigated the effects of this temporal alignment on university students' technology acceptance, self-efficacy, and cognitive load. A sequential explanatory mixed-methods design was employed with 150 university students. In the quantitative phase, monochronic and polychronic learners were assigned to interact with either a personalized interface matched to their time orientation or a universal interface serving as a control. Neutral learners were assigned to the universal interface to assess its baseline effectiveness. Pre- and post-tests were used to measure the key variables. Subsequently, semi-structured interviews with 30 participants provided qualitative insights into their experiences. Quantitative results revealed that, for both monochronic and polychronic learners, using a temporally aligned interface led to significantly higher technology acceptance and self-efficacy, as well as significantly lower cognitive load, compared to the universal baseline. Qualitative analysis confirmed these findings, showing monochronic learners valued structured, sequential workflows while polychronic learners thrived in flexible, multitasking environments. The universal interface was found to be effective and cognitively manageable for neutral learners. This study makes a significant contribution by establishing time orientation as a critical, empirically validated dimension for personalization in m-learning. Theoretically, it extends Person-Environment Fit theory to the design of digital learning environments. Practically, it provides a novel framework (TOPI) with actionable principles that challenge the dominance of one-size-fits-all design by demonstrating a clear path toward creating more effective, adaptive, and user-centered learning systems. This study provides compelling evidence that while a generic interface may be adequate for neutral learners, it is suboptimal for those with stronger temporal dispositions. Designing adaptive interfaces that accommodate temporal diversity is a key strategy for enhancing learning quality beyond mere usability, offering a practical framework for creating more effective and engaging digital learning environments.
Keywords:	Time Orientation; Personalized Interface Design; Educational Technology; Mobile Learning; Cognitive Load; Person-Environment Fit; Adaptive Learning.
DOI:	https://doi.org/10.5281/zenodo.18822693
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	END-TO-END SECURE COMMUNICATION IN FOR WIRELESS MULTIMEDIA SENSOR NETWORKS VIA MODIFIED GORILLA TROOPS OPTIMIZER DRIVEN DATA COMPRESSION WITH ENCRYPTION APPROACH
Author:	K. VINAYAKAN, V. ALAMELU MANGAYARKARASI, Dr. A. DINESH KUMAR, M. VASUKI, R. JAYAKUMAR
Abstract:	Wireless Multimedia Sensor Networks (WMSNs) have shifted the focus from traditional scalar wireless sensor networks (WSNs) to networks equipped with multimedia devices capable of capturing images, audio, video, and scalar sensor data. WMSN can provide multimedia content owing to the availability of low-cost CMOS microphones and cameras in addition to the considerable progress in multimedia source coding and distributed signal processing techniques. In comparison to a compressed image, an uncompressed image transmission consumes more energy, and, as a result, it becomes a necessity to establish an energy-aware compression technique to prolong the sensor node and the network lifetime. Simultaneously, security encryption systems can be used to protect the data being transferred, which ensures integrity and confidentiality against tampering and unauthorized access. This study develops an Integrated Approach for Secure and Energy-Efficient Data Transmission in Wireless Multimedia Sensor Networks (IASEE-DTWMSN) technique. The purpose of the IASEE-DTWSN method is to ensure security and maximum energy efficiency in WMSN via data compression and encryption algorithms. Initially, the IASEE-DTWSN technique compresses the images captured by WMSN using a discrete cosine transform (DCT) approach. Besides, the secure transmission of the compressed data can be accomplished using an advanced encryption standard (AES) model. Furthermore, the IASEE-DTWSN technique involves the design of a Modified Gorilla Troops Optimizer (MGTO) for optimizing the DCT and AES models. The MGTO approach is applied for determining the optimum quantization parameters of the DCT methodology and encryption key selection of the AES model in such a way that the compression ratio and PSNR are maximized. To validate the performance of the IASEE-DTWSN methodology, a broad array of simulations was involved. The experimental outcomes inferred that the IASEE-DTWSN model resulted in enhanced energy efficiency and security in the WMSN.
Keywords:	Wireless Multimedia Sensor Networks; Advanced Encryption Standard; Gorilla Troops Optimizer; Data Compression; Encryption
DOI:	https://doi.org/10.5281/zenodo.18822704
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	EXPLORING AI-GENERATED VISUAL THEMES IN ASIAN ART AS A FOUNDATION FOR ADVANCING DIGITAL ART RESEARCH
Author:	DEVI YURISCA BERNANDA , FILSCHA NURPRIHATIN , JOHANES FERNANDES ANDRY , FRANCKA SAKTI LEE , ANGIE WIYANI PUTRI , RION COMERON
Abstract:	This study aims to analyze emerging trends, and visual themes present in AI generated artworks across Asia by employing Descriptive, Predictive, and Prescriptive Analytics using the Tableau platform. The dataset utilized in this research comprises 10,000 AI generated image entries, focusing specifically on attributes such as digital art genre, production time, and Asian regional context. Through comprehensive analytical stages, the study examines how AI models construct visual styles, reinterpret artistic traditions, and synthesize cultural elements within the broader landscape of contemporary digital art. The primary challenge addressed in this research concerns the difficulty of identifying promising digital art genres for further exploration. This complexity arises from the increasingly diverse stylistic variations, the continuous evolution of visual themes, and the limitations in categorizing AI generated artworks that blend multiple genres or introduce novel styles that lack clear definition. To address these challenges, the methodological framework consists of four major stages: data acquisition and preprocessing using publicly available datasets, feature extraction and exploratory data analysis to uncover meaningful patterns and genre distributions, predictive and prescriptive analytics to determine potential genre trajectories and provide actionable recommendations and validation and interpretation to ensure analytical robustness and contextual relevance. The findings highlight the importance of sustained thematic innovation, deeper exploration of visual styles, and deliberate integration of localized cultural motifs to ensure that AI generated art remains relevant, competitive, and culturally grounded within Asian markets. Furthermore, strategic recommendations are formulated for each identified genre category to enhance artistic quality and strengthen the market positioning of AI generated digital artworks. This research is expected to contribute significantly to the advancement of AI driven artistic production in Asia by offering insights that are technical, aesthetic, and cultural in nature, ultimately supporting the development of more informed, sustainable, and contextually responsive digital art practices.
Keywords:	Artificial Intelligence, Digital Art, Dataset, Predictive Analytics, Art Genres
DOI:	https://doi.org/10.5281/zenodo.18822713
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	REAL-TIME INTELLIGENT CLAIM VERIFICATION USING BLOCKCHAIN-BASED RPA AND XGBOOST FOR SCALABLE HEALTHCARE FRAUD DETECTION
Author:	DR.V. BALASANKAR, DR. V SRINADH, DESIDI NARSIMHA REDDY, JALA PRASADARAO, K SWETHA, ELANGOVAN MUNIYANDY
Abstract:	Healthcare insurance fraud has been a major problem and requires smart, scalable, and auditable detection systems. The study presents a framework of Robotic Process Automation (RPA) with Integrated Blockchain technology and integrates the methods of detecting fraudulent claims in real-time with Machine Learning (ML) and RPA. The system is developed in five layers, including data input, automation, intelligence, smart contracts, and blockchain ledger. It works on a structured dataset of 4,000 healthcare claims having 83 attributes. Claim intake and rule-based screening are automated by RPA bots, and fraud classification against predefined risk thresholds is done by an ML engine driven by extreme gradient boosting using smart contracts, with final results being impartially stored by Hyperledger Fabric. The proposed framework significantly outperformed the traditional models of predicting data (Convolutional Neural Network, Recurrent Neural Network, Decision Trees, and Logistic Regression) with a result of 92.4 percent of classification accuracy. It guarantees scalability and traceability without compromising the latency by making claims in under 2.5 seconds. Measurements of feature importance, e.g. KullbackLeibler Divergence, Entropy, and Gini, are more explainable and guide automation logic. What this hybrid framework will achieve is a new standard of real-time, audible, and intelligent fraud detection in healthcare insurance systems that will provide a technically sound and legally compliant basis of foundation to next-generation claims processing.
Keywords:	Healthcare Insurance, Blockchain, Fraudulent Claims, Hyperledger Fabric, Robotic Process Automation.
DOI:	https://doi.org/10.5281/zenodo.18822727
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	EFFINET-CL: A FRAMEWORK FOR LEAF DISEASE DETECTION AND CLASSIFICATION USING CONTINUAL LEARNING
Author:	SREYA JOHN, Dr. P. J. ARUL LEENA ROSE
Abstract:	This paper presents an innovative approach to detect and classify leaf diseases using EfficientNetB3 combined with continual learning strategies. The proposed model leverages data augmentation techniques for improved generalization and integrates continual learning through Elastic Weight Consolidation (EWC) and Replay Memory to prevent catastrophic forgetting. This approach keeps the model's knowledge of formerly learnt diseases intact while permitting it to learn continually from new disease classes. The integration of an Out-of-Distribution (OOD) detection mechanism further improves the ability of the model to recognize unknown diseases, making it a reliable tool for real-time agricultural applications. The model demonstrated excellent performance with an accuracy of 98.44%, highlighting the potential of deep learning combined with continual learning for plant disease diagnosis, and outperformed ResNet50, DenseNet121, InceptionV3, and MobileNet-V2 algorithms.
Keywords:	Leaf Disease Detection, Feature Extraction, Classification, Deep Learning, Continual Learning
DOI:	https://doi.org/10.5281/zenodo.18822739
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	SOCIAL NETWORKS AS A COMMUNICATION ENVIRONMENT FOR YOUNG PEOPLE: INFORMATION THREATS AND EUROPEAN APPROACHES
Author:	LIUDMYLA MIALKOVSKA, IRYNA ZABIIAKA, VITA STERNICHUK, LARYSA PYLYPIUK, YULIIA LITKOVYCH, LIUDMYLA DOLOZHEVSKA
Abstract:	The aim of the study is to identify linguistic tools for creating information risks in the interpersonal interaction of young people in the field of social networks and to find ways to develop adjusted countermeasures based on the European experience in the field of media literacy. The study was characterized by an interdisciplinary strategy that combined typological analysis of linguistic phenomena, mechanistic representation of cognitive processes, and comparative analysis of European media education practices. The research is based on the synthesis of forty-six scientific publications, archives of fact-checking organizations, and official documents of the European Commission. Еhe three most serious risk profiles among young people were identified: emotional and psychological, cognitive and cultural, and digital and legal. It was found that the combination of emojis with youth slang constitutes the highest risk index, reaching a value of 0.92 conventional units. By increasing the level of intra-sentence code-switching by 34 percent, young people tend to believe highly questionable sources more than neutral messages. TikTok video memes proved to be the most effective in spreading disinformation among young people, reaching a score of 8.1 conventional units due to a combination of 96% spread rate and only 28% critical reflection on the content. Adolescents aged 13-16, who have an emotionally impulsive personality type and low digital literacy, demonstrate a total risk of 7.8 conventional units and are the most vulnerable group.ь The effectiveness of such an adaptation as the model of European media literacy strategies implemented in Ukraine has the potential to be 69.8-85.1% of international standards. The results of the study are used to formulate a theoretical framework focused on the development of individual educational activities aimed at combating disinformation, which would take into account the communicative behavior of adolescents in the bilingual sphere of the Ukrainian population.
Keywords:	Internet Communication, Social Networks, Normative Communication, English, Ukrainian, Social Networks, Stylistics, Memes, Emojis, Abbreviations
DOI:	https://doi.org/10.5281/zenodo.18822757
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	AN INTELLIGENT PRIORITY-BASED TELEMEDICINE ALGORITHM FOR REAL-TIME EMERGENCY HEALTH MANAGEMENT IN MILITARY ENVIRONMENTS
Author:	DR.S. HEMALATHA, DR. VIJAYA KUMBHAR, SURYA LAKSHMI KANTHAM VINTI, K. PADMAPRIYA, DR.S. MURUGANANDAM, S. K. SATYANARAYANA
Abstract:	In military environment especially in the battlefield duration the timely medical treatment is saving life, but the conventional telemedicine scheme is relay on first-come, first-served (FCFS) scheduling which causes the failed treatment for the emergency patient. To overcome such a scenario many other methods were proposed with latest technologies still the need of novel method to save such a people. This article is address the same , by proposing the novel algorithm proposes an Intelligent Priority-Based Telemedicine Algorithm (IPTA) which is assumed as a wearable device with the modern technologies of sensors, Internet of Thing and Machine learning to predict the soldiers vital parameters such as heart rate, oxygen saturation, blood pressure, and body temperature are continuously monitored and analyzed through machine learning–based predictive analytics to estimate each soldier’s health criticality. This proposed framework automatically identified the soldiers body condition and assign the priority to meet the telemedicine doctor. This experiment was tested using the simulation tools of MATLAB using Python and OMNET++ for connectivity related Quantitative and qualitative metric comparison with the traditional methos of FCFS, Rule Based and Fuzzy logic methods. Quantitative results expose proposed IPTA framework shown the improvements in accuracy (91.6%), resource utilization (91%), and packet delivery ratio (94%), latency (85 ms) and average response time (2.9 s). Also, system throughput increased by 31%, and survival probability improved by 24%.
Keywords:	Telehealth, Internet of Things, Wearable Device, Telemedicine Algorithm, Artificial Intelligence
DOI:	https://doi.org/10.5281/zenodo.18822763
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	CAN FEDERATED DIFFERENTIALLY PRIVATE MULTI-MODAL TRANSFORMERS IMPROVE RARE HEALTHCARE FRAUD DETECTION?
Author:	VISHAL NAMIREDDY, DR. DURAIRAJ.M, DESIDI NARSIMHA REDDY, UDAYALAXMI ADITYA TEKI, ELANGOVAN MUNIYANDY, BALAKRISHNA BANGARU
Abstract:	An ongoing challenge in healthcare insurance fraud detection is designing intelligent algorithms that identify subtle and evolving fraud patterns while preserving patient privacy. We propose FraudBERT-MM, a multi-modal framework that fuses structured claim features with unstructured textual explanations to form comprehensive claim representations. The architecture combines a BERT-based text encoder, an MLP-based structured encoder, a cross-modal fusion layer and a binary classification head. To address scarce labelled fraud examples, we apply prototype-based few-shot learning with episodic training. For privacy-preserving collaboration, models are trained in a federated learning setup augmented with Gaussian differential privacy, achieving a privacy budget of ε = 1.87. Unlike prior work that assumes large centralized labeled datasets and single-modality inputs, FraudBERT-MM integrates multi-modal learning, few-shot adaptation and formal privacy guarantees in one unified pipeline. We evaluate the approach on a real-world dataset of 4,000 claims with 83 features. FraudBERT-MM attains an F1 score of 0.882, outperforming centralized non-private, text-only and structured-only baselines. The model also converges 26% faster than the centralized baseline while preserving strong privacy, with only a small utility loss from noise. Experimental results indicate improved detection of rare and emerging fraud trends without compromising data privacy. We provide implementation details, ablation studies and deployment guidelines to support reproducibility and accelerate safe adoption in clinical and insurance settings across diverse operational contexts. These findings suggest that combining multi-modal representations, few-shot adaptation and federated differential privacy offers a practical and effective path toward deployable, privacy-aware healthcare fraud detection systems.
Keywords:	DDoS Detection, Contrastive Learning, Large Language Models (LLM), Firewall, Hybrid Model.
DOI:	https://doi.org/10.5281/zenodo.18822773
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	ARTIFICIAL INTELLIGENCE-BASED EARLY PREDICTION OF THERMAL RUNAWAY IN AGV BATTERY CELLS
Author:	OH CHANG-HO, MA TAE-YOUNG, KIM YONG-KUK, GEUN-JIN AHN, PARK SUNGBUM
Abstract:	Automated Guided Vehicles (AGVs) in semiconductor manufacturing operate continuously with lithium-ion battery packs which undergo thermal runaway under abusive conditions such as over-charge, over-discharge or cooling failure. This study presents an artificial-intelligence (AI) - based early - warning framework that predicts imminent thermal runaway in AGV battery cells. We collected a large-scale dataset of 12.7 million time-series records from the battery-management system (BMS) of AGVs in the real factory environment, including cell-level temperature, voltage, current, alarms and cycle counts information. Three predictive models — XGBoost, 1D CNN and LSTM — were developed to forecast the next-step maximum cell temperature using a 24-minute sliding window. LSTM achieved the best performance (MAPE ≈ 1.66 %) compared to CNN (≈ 1.67 %) and XGBoost (≈ 1.76 %). Based on LSTM residuals and predicted threshold exceedance, an early warning mechanism was implemented, raising alarms several minutes before the temperature crossed critical limits. The results demonstrate that AI-driven forecast algorithm enables practical pre-runaway warning and can be embedded in AGV safety-monitoring systems in high-throughput semiconductor logistics applications.
Keywords:	Thermal runaway, Lithium-ion battery, Automated Guided Vehicle (AGV), LSTM, CNN, XGBoost, Early warning, Battery management system (BMS)
DOI:	https://doi.org/10.5281/zenodo.18822793
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	PERSONALIZED RESEARCH RECOMMENDATION BASED ON TEMPORAL FUSION TRANSFORMER FOR AMAZON FASHION PRODUCTS
Author:	GOKILAVANI A, DR P AMUDHA
Abstract:	Personalized product recommendations have become more valuable in recent years as a means of improving the online shopping experience for customers. However, accurate sales forecasting for fashion products remains challenging due to high demand variability, short product life cycles, and complex temporal dependencies influenced by pricing, discounts, and customer behavior. A robust framework for sales prediction is created using a structured pipeline that integrates data collection, preprocessing, feature selection, predictive modeling, and performance analysis. The first step in the inquiry is gathering sales data from Amazon, a dataset known for its size, complexity, and range of features. To ensure data quality and relevance, data is put through preparation processes like data cleaning, transformation, and exploratory data analysis (EDA), which help to identify patterns and correlations in the dataset and prepare it for modeling. During the feature selection stage, the Boruta algorithm is employed, utilizing its ability to identify significant characteristics that improve the model's predictive capabilities. This approach retains only the most relevant information, reducing computing complexity and increasing model efficiency. For the prediction phase, three different models are examined: a CNN-BiLSTM model, which combines convolutional layers for feature extraction with bidirectional LSTMs for sequential learning; a Temporal Fusion Transformer (TFT), which is well-known for its interpretability and ability to control temporal dynamics; and a hybrid model that incorporates aspects of both CNN and TFT. This combination aims to capture complex temporal patterns and interactions within the data to optimize forecast accuracy.The models' performance is evaluated using a variety of error metrics, including Mean Absolute Percentage Error (MAPE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Weighted Mean Absolute Percentage Error (WMAPE). These measures provide a comprehensive picture of prediction accuracy by highlighting both absolute and relative errors. By using a range of evaluation criteria, this study ensures that the models' performance is looked at from multiple perspectives, offering insights into the models' practicality and robustness. Experimental results show that the proposed hybrid CNN-TFT model outperforms CNN-BiLSTM and standalone TFT models, achieving an accuracy of 97.25% with lower MAPE (0.383) and RMSE (0.1908). the results demonstrate that the proposed hybrid framework provides reliable and accurate sales forecasts, supporting effective demand planning, inventory optimization, and decision-making in fashion e-commerce platforms.
Keywords:	Temporal Fusion Transformer, CNN, LSTM, Fashion Sales, Prediction.
DOI:	https://doi.org/10.5281/zenodo.18822814
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	SIMPLIFIED ROUTING FOR MOBILE AD HOC NETWORKS USING A COMBINED RESOURCE METRIC
Author:	MOHAMMED ABDUL BARI, ABDUL WASE MOHAMMED, ARSHAD AHMAD KHAN MOHAMMAD, SIREESHA MOTURI, ANUSHA MAROUTHU
Abstract:	Mobile Ad Hoc Networks are autonomous, infrastructure-less networks consisting of wireless mobile devices. Within the network, “the sender communicates with the receiver only if it is in radio communication range of one another; otherwise, it must communicate through intermediate nodes. To enable reliable communication, cooperation among intermediate nodes and the use of their resources are essential; otherwise, packets are dropped. The primary reasons for these drops are “battery drain,” “full buffers,” and “high traffic at the intermediate nodes.” These packet drops are unintentional. This paper proposes a simplified routing method to address undesirable packet drops by planning routes that avoid congestion and maximize resource utilization at nodes. These routes are determined based on each intermediate node's remaining power, available memory, and current traffic levels. Performance evaluation of this protocol is conducted on a simulated network and compared with other standard reactive methods, accounting for uniform traffic distribution. The simulation results show that this protocol can significantly reduce undesirable packet drops and increase successful transmissions, ensuring a reliable communication system for critical applications, including disaster relief operations and healthcare systems.
Keywords:	Mobile Ad Hoc Networks, Packet Drops, Routing, energy, buffer, Packet Delivery.
DOI:	https://doi.org/10.5281/zenodo.18822819
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	AN INTEGRATED ENHANCED DEEP LEARNING MODEL DETECTING AND CLASSIFYING CROP DISEASES IN THE EARLY STAGES TO MAXIMIZE THE YIELD
Author:	DHIRAJ BHISE, SUNIL KUMAR, HITESH MOHOPATRA, BHUPESH KUMAR SINGH
Abstract:	Agriculture sector is the most promising sector which plays crucial role in Economy data of every country. There are numerous challenges are being faced by the farmers. Hence, researchers and academicians are trying to overcome these challenges by introducing new techniques, machine learning and AI models. Among the many challenges faced by farmers, leaf diseases stand as formidable adversaries, persistently diminishing their harvests. In most cases, farmers’ harvests are diminished by leaf diseases which directly affect the crop yield. Researchers and academicians are developing a generalized model applicable to all the crops. In the initial stage we have focused on wheat crop. Wheat is cultivated at large and is consumed by maximum people to its status as a staple food. Similarly in the next subsequent development of generalized model we will focus on many other crops. Wheat plant diseases affect the yield in the recent decades. Decreased crop yield directly affects the economic growth of the country and have social impact on farming. Recent technological advancements using artificial intelligence (AI), machine learning (ML) and deep learning (DL) have made it easier to identify diseases on leaves in their early stages. With the help of recent technologies it became possible to develop ‘autonomous’ systems keeping eye on the ‘diseases’ in the early stages. In this approach, we have used an open dataset from Plant Village which has more than images of 5000 plant leaf diseases divided into 04 types of leaf diseases along with the normal leaf. The proposed system is a combination of CNN and Mask RCNN. In CNN method we have successfully identified the type of disease for the input image whereas Mask RCNN masks the exact area covered by the disease of the leaf. Proposed integrated deep learning model applies the convolutional neural networks (CNN) for better accuracy (97 % accuracy), classification, and identifying the disease affected region. In the proposed work, model integrates 02 different methods CNN and Mask RCNN.
Keywords:	Deep Learning, Machine Learning, Region Based Convolutional Neural Network, Leaf Disease; Artificial Intelligence, Classification Model, Forecast Illnesses.
DOI:	https://doi.org/10.5281/zenodo.18823735
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	EDGE-CENTRIC PRIVACY-PRESERVING VIDEO ANALYTICS FOR LATENCY-SENSITIVE MONITORING
Author:	DR GAURAV VISHNU LONDHE, DR. S. MAHESWARI, DR.V.RADHIKA, LAVANYA KUMARI PITHANI, V SIVARAMARAJU VETUKURI, ELANGOVAN MUNIYANDY
Abstract:	Real-time video analytics at the network edge is essential for smart surveillance, autonomous monitoring, and safety-critical decision making. However, maintaining high detection accuracy under severe latency limits and demonstrable privacy protection is difficult. Raw video transmission on the cloud increases delay and privacy risks, while entirely on-device solutions are limited by computational and energy resources.This paper presents a hybrid edge-centric video analytics architecture that balances latency, accuracy, and privacy via split inference, lightweight on-device processing, and privacy-preserving feature preservation. A compact backbone's early layers do motion filtering, region-of-interest extraction, and feature computing on the device, while a nearby edge server performs deeper inference. Quantization and differential privacy safeguard intermediate characteristics, with selective homomorphic encryption applied to tiny crucial layers for better guarantees. Based on bandwidth and device demand, an adaptive runtime controller chooses the split point and compression level. On VIRAT, PEViD, and Sultani-derived datasets, the suggested technique achieves sub-100 ms median end-to-end latency and near-cloud detection accuracy. Results indicate that modest differential privacy (ε ≈ 1) significantly reduces membership-inference attack success with minimal accuracy loss, whereas full homomorphic encryption is too latency-intensive. Split inference and lightweight differential privacy create a realistic and deployable privacy-aware real-time edge video analytics solution.
Keywords:	Video Analytics, Split Inference, Differential Privacy, Homomorphic Encryption, Latency Constraints, Adaptive Controller
DOI:	https://doi.org/10.5281/zenodo.18823749
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	QUANTUM-ENHANCED FEATURE EXTRACTION (QUFEX): A CONCEPTUAL FRAMEWORK FOR PRECISION ATMOSPHERIC RIVER ANALYSIS
Author:	SIVACHITRALAKSHMI , CHITRA P
Abstract:	Atmospheric Rivers (ARs) are major contributors to global hydrological variability and flood hazards, particularly in regions influenced by monsoonal systems. Accurate detection of ARs remains a challenge due to their dynamic, non-linear spatial patterns. This study presents a comparative evaluation of three advanced deep learning models—U-Net, CGNet, and DeepLabV3+—and proposes Q-ARNet, a quantum-enhanced atmospheric river detection framework based on a Quantum Neural Network (QNN) for image segmentation. The experiments were conducted on a dataset comprising 574 AR events from 1951 to 2020, incorporating variables such as integrated water vapor (IWV), zonal and meridional wind components (U850, V850), and sea-level pressure (PSL). Q-ARNet outperforms current classical benchmarks by approximately 3.5% in accuracy while utilizing a substantially reduced input feature set, thereby addressing the computational bottleneck associated with high-dimensional climate data. The proposed framework achieves an average accuracy of 97.5 ± 0.3%, an Intersection over Union (IoU) of 94.5 ± 0.4%, and an F1-score of 95.7 ± 0.2%, consistently surpassing classical deep learning models. These findings highlight the potential of quantum neural networks to enhance both the representational capacity and computational efficiency of atmospheric modeling through quantum parallelism and entanglement. The study contributes to the growing intersection of quantum computing and climate informatics, offering a promising framework for improved flood forecasting and climate risk management.
Keywords:	Atmospheric Rivers, Flood Forecasting, Deep Learning, Quantum Neural Networks, Semantic Segmentation, Evaluation Metrics.
DOI:	https://doi.org/10.5281/zenodo.18823768
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	EFFICIENT MPPT FOR PV SYSTEMS UNDER SHADING CONDITION USING HYBRID METAHEURISTIC OPTIMIZATION MODEL
Author:	HARITHA INAPAGOLLA , R ASHOK BAKKIYARAJ, KATRAGADDA SWARNA SRI
Abstract:	The increasing penetration of photovoltaic (PV) systems in the energy generation mix make it necessary for Maximum Power Point Tracking (MPPT) techniques to be more and more efficient. Solar (PV) systems are generally used under partial shading and operate at many local peaks on the power-voltage (P-V) curve hence making Global Maximum Power Point (GMPP) tracking hard. In such conditions, traditional MPPT methods like Perturb and Observe (P&O) as well as Incremental Conductance (IC), most of the time fails to track GMPP effectively, leading towards higher power losses resulting in decreased efficiency. In this research a new hybrid metaheuristic method, which combines Particle Swarm Optimization (PSO) and Differential Evolution (DE), is suggested for improving MPPT performance in PV systems. With the help of hybrid PSO-DE model, which has better convergence speed and exploration property as well as robustness to local minimum with random search directions (Differential Evolution), it helps locate GMPP correctly even under various shading situations. The work provides novelty by suggesting an organized hybrid PSO-DE based MPPT framework that strictly weighs the quick convergence and worldly search to dependable Global Maximum Power Point tracking to track the partially shaded conditions. The proposed method incorporates both PSO exploitation and DE exploration in one optimization loop, which makes it more stable, converges faster, and is more resilient than current standalone or weakly connected hybrid MPPT techniques. The results clearly indicate substantial improvements in the tracking speed, power output and stability with respect to existing methods when the proposed hybrid PSO-DE method is employed especially for not uniform set points. The results of this study can help in improving the efficiency and reliability of solar energy systems so that PSO-DE hybrid model usable for real world applications. This work advances the field of MPPT by supporting new technologies to respond to a fluctuating set behaviour for renewable energy systems and provides insight into more efficient solar solutions.
Keywords:	Photovoltaic Systems, Maximum Power Point Tracking, Partial Shading, Global Maximum Power Point, Hybrid Meta-heuristics, Particle Swarm Optimization, Differential Evolution.
DOI:	https://doi.org/10.5281/zenodo.18823775
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	ONTOLOGY-AWARE MULTIMODAL NLP FRAMEWORK INTEGRATING TRANSFORMER–CNN FUSION FOR ENHANCED MEDICAL DATA AND MANUSCRIPT ANALYSIS
Author:	EMMADI SUSHMA , Dr. SUKLA SATAPATHY
Abstract:	The exponential growth of unstructured medical data—ranging from clinical notes and electronic health records (EHRs) to academic publications—creates substantial difficulties for automated interpretation. Conventional natural language processing (NLP) models, though useful for textual information, are limited in their ability to assimilate additional modalities such as diagnostic imaging, graphical records, and clinical diagrams, which are essential for a holistic analysis. To address this limitation, we present an ontology-guided multimodal NLP framework that integrates Transformer-based encoders for textual inputs with convolutional neural networks (CNNs) designed for medical images and diagrams. A specialized fusion layer combines these heterogeneous representations, thereby enabling context-sensitive reasoning and predictive modeling. The framework is further enhanced through the incorporation of established medical ontologies, including UMLS and SNOMED CT, which improve semantic accuracy and interpretation of domain-specific terminology. For empirical validation, the system was tested using two benchmark datasets: MIMIC-III, which provides comprehensive textual medical records, and Open-i, which offers a large collection of biomedical images. The proposed framework consistently outperformed leading baselines, achieving 97.36% accuracy, 96.82% precision, 96.59% recall, and a 96.70% F1-score. In addition to strong predictive performance, the architecture demonstrates scalability, interpretability, and compliance with stringent privacy standards, making it practical for deployment in real-world healthcare environments. By effectively unifying multimodal information with domain-specific ontological knowledge, this work delivers a computationally efficient and clinically relevant solution for advancing automated medical document analysis and decision support.
Keywords:	Natural Language Processing; Deep Learning; Multimodal Data Integration; Transformer Networks; Convolutional Neural Networks; Medical Ontologies; Medical Informatics; Clinical
DOI:	https://doi.org/10.5281/zenodo.18823802
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	FE-DADT: A FEATURE ENGINEERING DRIVEN DDOS ATTACK DETECTION TECHNIQUE AND EVALUATION FOR EFFECTIVE MITIGATION
Author:	SWATHI KATHI, NARSIMHA GUGULOTHU
Abstract:	Distributed denial-of-service (DDoS) attacks have been a persistent threat to the continuous operation of network services by depleting its essential resources and normal users cannot access the services. With the increasing evolution of these attacks, simple detection of differentiating malicious packets from innocent packets within traffic has become a complex task. Therefore, effective mitigation relies on accurate identification of discriminatory features of the network as opposed to similar (but non-redundant) sample-level analysis in higher dimensional space. In this paper, we present a filter-based feature selection method rooted in the statistical validity of the Student t-test that identifies the strongest predictors from traffic flow data. The task is to find a minimal but optimal set of features that improves classifier performance by noise and redundancy removal. On CICDDoS2019 benchmark dataset, the model was trained with 399,998 and tested on 112,611 samples. The results show that t-test is able to detect 58 features with statistical significace. With this optimized feature set, a eXtended Gradient Boosting (XGBoost) classifier reached a 99.82% detection rate and 97.40% classification accuracy. Our results show that a feature elimination approach based on statistical grounds, strengthens a significant increase in the performance of DDoS detection systems based on ML methods.
Keywords:	DDoS, Classification, Network traffic flow, Prediction, Feature selection, Hypothesis Testing, Machine Learning
DOI:	https://doi.org/10.5281/zenodo.18823810
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	FEDERATED LEARNING WITH BLOCKCHAIN FOR SECURE AND SCALABLE FINANCIAL SERVICES
Author:	DR. NAIM SHAIKH, DR. A.PANKAJAM, DR. VIVEK VEERAIAH, SHEETAL PRADIP PATIL, DR. MAMATHA G, DR. SRIDEVI.R, ANKUR GUPTA, DR. M.YELLAIAH NAYUDU
Abstract:	People are justifiably apprehensive about the safety, privacy, and scalability of their data in collaborative machine learning contexts now that digital financial services are available. But it is hard to get to, and different groups don't trust it. Blockchain is a kind of distributed ledger technology that stores and checks data in a way that doesn't need a trusted third party. However, when used alone, it has problems with its size and battery use. This research puts out a hybrid architecture that combines blockchain with FL to solve these problems and develop financial services that can grow. This approach, which uses blockchain technology, protects the data sovereignty of organisations. It also makes it easier for financial organisations to work together to build global models. It does this by being able to keep an eye on changes, build trust, and provide people the chance to get personal incentives. The suggested strategy is better than FL-only and blockchain-only solutions when it comes to accuracy, scalability, and safety. The amount of energy used and the time it takes to respond to queries are also maintained at a reasonable level. Research can perform a lot of important things with the system, including look into claims of fraud, assess credit risk, and stop money laundering. The outcomes help make the financial industry's digital infrastructure safer, more open, and more in line with the law.
Keywords:	Federated Learning, Blockchain, Financial Services, Security, Fraud Detection, Credit Risk Analysis, Anti-Money Laundering.
DOI:	https://doi.org/10.5281/zenodo.18823826
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	INNOVATIVE APPROACHES TO EDUCATIONAL KNOWLEDGE GRAPHS: LEVERAGING NEO4J AND LLM FOR SYNTHETIC DATA GENERATION
Author:	TANTY OKTAVIA , M BAGASKORO TRIWICAKSANA S
Abstract:	In todays rapidly evolving higher education landscape, the utilization of data is critical for gaining insights that enhance institutional effectiveness. However, accessing and leveraging this data presents significant challenges due to privacy, licensing, and security constraints. This research evaluates the effectiveness of using Large Language Model (LLM)-generated synthetic data for constructing and visualizing educational knowledge graphs (KGs) using Neo4j. By integrating Neo4j's graph capabilities with LLM's generative power, we developed comprehensive knowledge graphs across five key university domains: curriculum development, personalized learning, learning analytics, resource management, and alumni engagement. Our methodology employs a two-stage prompting strategy with Claude 3.5 Sonnet to generate synthetic datasets that simulate real-world educational scenarios, achieving data completeness rates of 77.55-100% and domain alignment scores of 82.84-100%. The synthetic data successfully captured complex educational relationships through varied node types (5-16) and relationship types (4-16) across domains, while relationship accuracy metrics (38.21-46.55%) demonstrated focused, meaningful connections. Implementation results show efficient processing times (3-8 seconds per domain) and strong practical utility in modelling educational patterns and relationships. This approach significantly reduces data collection time and resources while maintaining data privacy and integrity. While limitations exist regarding validation against real-world datasets, our findings demonstrate that LLM-generated synthetic data can effectively prototype educational knowledge graphs, offering institutions a scalable, privacy-preserving framework for developing data-driven insights and decision-making capabilities.
Keywords:	Knowledge Graphs (KGs), Neo4j, Synthetic Data, Large Language Models (LLMs), Curriculum Development, Personalized Learning, Learning Analytics, Resource Management, Alumni Engagement
DOI:	https://doi.org/10.5281/zenodo.18823853
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	A BLOCKCHAIN-DRIVEN MULTI-AGENT FRAMEWORK FOR ENHANCING TRANSPARENCY, EFFICIENCY, AND RESILIENCE IN SMART SUPPLY CHAIN MANAGEMENT
Author:	DR. P. SANTHOSH KUMAR, DR. A.PANKAJAM, DR. VIVEK VEERAIAH, DR. ADVETA GHARAT, DR. MAMATHA G, DR. DEVIKA RANI ROY, ANKUR GUPTA, N RAJITHA
Abstract:	As global supply chains have evolved quickly, challenges including transparency, collaboration, security, and the capacity to make choices in real time have become increasingly important. The present research fails to offer a unified, extensible framework that integrates decentralised trust, intelligent optimisation, and resilience across diverse supply chain contexts, even though blockchain and multi-agent systems have been examined independently to enhance autonomy and traceability. Most of the research done before this one is either too small in scope or hasn't been able to prove that integrated blockchain-MAS systems work in real life. Also, most of the studies have only looked at single services, like traceability or finance. The present study addresses this informational gap by proposing a novel Blockchain-Driven Multi-Agent Framework for the optimisation of supply chain activities from inception to completion. This system combines decentralised ledger technology, autonomous agents, reinforcement learning, and smart contract automation. The framework brings new information to the table by adding a scalable on-chain/off-chain data strategy, smart techniques to optimise for cost, resilience, and sustainability, and formalising hybrid agent-based coordination on blockchain. In the agricultural, industrial, and logistical sectors, simulation-based evaluations show big improvements over traditional methods in the following areas: transparency (+65%), speed of decision-making (+48%), time to recover from an interruption (-60%), energy efficiency (+32%), and reduction of product spoilage (-66%). The research contributes to the current knowledge on intelligent, decentralised supply chains by offering a generalisable architecture, empirical performance data, and practical insights.
Keywords:	Blockchain, Multi-Agent Systems, Smart Supply Chain, Reinforcement Learning, Decentralized Optimization, Transparency, Resilience, Sustainability
DOI:	https://doi.org/10.5281/zenodo.18823863
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	FEDERATED DEEP LEARNING FOR DIABETES PREDICTION IN INDIA: ENHANCING PRIVACY AND INTERPRETABILITY WITH DP-SGD AND LIME
Author:	P. CHITRALINGAPPA , DR RAMATENKI SATEESH KUMAR , ARTIKA FARHANA, SHYAMSUNDER CHITTA , SRIDEVI GAMINI , SANDA SRI HARSHA
Abstract:	Diabetes prevalence in South Asia is increasing, and automated risk prediction is necessary to enable early intervention. This work proposes FedHybNet, a privacy-preserving federated hybrid architecture that combines a TabTransformer backbone with per-client personalization, applies differentially private stochastic gradient descent (DP-SGD) for formal Differential Privacy (DP) guarantees, and uses Local Interpretable Model-agnostic Explanations (LIME) for instance-level interpretability. FedHybNet protects client updates via DP-SGD while preserving predictive utility through a hybrid aggregation scheme and lightweight on-device personalization. Performance and interpretability were evaluated on a recent public South-Asia tabular benchmark (DiaBD, 5,288 records, 2025) and on a larger synthetic India cohort (15,000 records) designed to match national demographics and covariate distributions. Models were compared to XGBoost (centralized classical baseline) and an advanced federated baseline (FedDL). Experiments used stratified 5-fold cross-validation and reported AUROC, AUPRC, F1-macro, expected calibration error (ECE), computational cost (GFLOPs), training time, and privacy budget (ϵ,δ). FedHybNet achieved AUROC 0.89 and AUPRC 0.69 at ϵ=2.0, improving AUROC by 0.01–0.03 (≈1.1–3.5% relative) over baselines and reducing calibration error (ECE) by 28.9–41.8% relative. Training cost increased due to the Transformer encoder and DP overhead, but the privacy–utility trade-off favored FedHybNet versus non-DP FL. These results indicate FedHybNet is practical for privacy-aware, interpretable diabetes risk prediction in distributed clinical settings.
Keywords:	Federated learning, Differential privacy, DP-SGD, Diabetes prediction, LIME, Interpretability.
DOI:	https://doi.org/10.5281/zenodo.18824252
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	GRASSHOPPER OPTIMIZED VISION TRANSFORMER WITH ADAPTIVE TOKEN CALIBRATION DROPOUT SCALING QKV ATTENTION REFINEMENT IN COTTON LEAF DISEASE IDENTIFICATION
Author:	K. KARTHIGA , Dr. B. RAJDEEPA
Abstract:	Accurate identification of cotton leaf diseases remains a complex task due to high visual similarity among disease types, irregular lesion boundaries, background interference, and the early-stage subtlety of symptoms in field-captured images. Current literature lacks transformer-based disease identification frameworks that dynamically regulate attention, token relevance, and learning stability in response to lesion morphology. This study bridges that gap by introducing an optimization-driven attention control mechanism that generates new insights into lesion-focused transformer learning for precision agriculture. Traditional models and vanilla Vision Transformers struggle with fixed patch segmentation, redundant token processing, and diluted attention focus, leading to suboptimal classification. To address these challenges, this research proposes a novel model titled Grasshopper Optimized Vision Transformer (GO-ViT), engineered for lesion-aware feature refinement in cotton leaf disease identification. GO-ViT integrates swarm-based Grasshopper Optimization to dynamically configure patch embedding size, attention head scaling, dropout distribution, QKV matrix refinement, and entropy-based token pruning. This biologically inspired regulation emulates adaptive foraging patterns, ensuring high lesion focus and reduced attention waste. GO-ViT enhances inter-class separation, preserves spatial continuity, and minimizes false activation over non-symptomatic regions. Experimental evaluation across five performance metrics confirms GO-ViT’s robustness, with 94.491% Overall Detection Efficiency and 88.985% Balanced Class Correlation. The model demonstrates superior ability to isolate overlapping infections, suppress background noise, and deliver stable predictions across diverse cotton leaf conditions. GO-ViT stands as a reliable solution for scalable, high-precision disease classification in precision agriculture, combining visual attention learning with nature-driven optimization control.
Keywords:	Cotton Leaf Disease, Vision Transformer, Grasshopper Optimization, Token Pruning, Attention Calibration, Lesion Segmentation
DOI:	https://doi.org/10.5281/zenodo.18824266
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	THE DEPLOYMENT OF DIGITAL HR TECHNOLOGIES AS A TOOL FOR ENHANCING ENTERPRISE PRODUCTIVITY IN THE DIGITAL ECONOMY
Author:	SEVIL RAMAZANOVA , OLHA MIZINA , NATALIA NOVHORODSKA , ULIANA KOROLOVA , DENYS KOROLOV
Abstract:	Relevance of the study The relevance of this study stems from the imperative to integrate HRTech solutions to enhance productivity within the digital economy, signifying a paradigm shift in human capital management, and the necessity to institutionalize Employer Branding as a driver for sustainable development. Purpose of the study The study aims at formalizing a hybrid HRTech Framework integrated with Employer Branding, alongside verifying its impact on labor productivity through econometric modeling, cognitive-emotional stratification, as well as architectural interoperability. Research methods The methodologies employed in this research include SWOT analysis, econometric modeling, structural synthesis frameworks, cognitive-emotional modeling, and econometric validation. Results obtained The integrative analysis revealed the superior efficacy of People Analytics (0.84; ΔLP +18.2%; fluctuation –25.3%), AI communications (0.82; TTTHR –21.4%; inclusivity +15.9%), and HRIS (0.80; HRΔCOE –23.1%; adaptability +39.6%). Based on these findings, a tri-layered HRTech Framework incorporating Employer Branding architecture was devised, which facilitates cognitive traceability and brand extrapolation. Upon validating the project framework, a noteworthy enhancement in key performance indicators was documented (ΔLP +19%, TTTHR –24%, RRI = 0.84, AIU = 0.88, EER = 0.86), indicating a high degree of cognitive relevance, HR conversion efficiency, and digital adaptability of the model within the context of strategic HR digitalization. Scientific novelty of the study The scientific novelty of this study lies in formalizing the HRTech Framework with Employer Branding as a hybrid architecture that synergizes People Analytics, AI communications, and HCM solutions with brand-oriented modules, thereby providing cognitive traceability, integrative influence, and affective validity in the context of digital productivity. Prospects for future research The prospects for future research entail comprehensive implementation of the HRTech Framework with Employer Branding in a tangible HR environment, concentrating on behavioral dynamics, institutional interoperability, and brand-oriented adaptation. It is advisable to initiate a controlled pilot project with traced verification of procedural resilience, AIU stability, as well as the framework’s digital efficacy.
Keywords:	Employment, Labour, Wage, Economic Empowerment, Smes, Sustainable Development, Green Job
DOI:	https://doi.org/10.5281/zenodo.18824280
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	PRIVACY-PRESERVING CUSTOMER DATA MANAGEMENT USING HYBRID AI-CRYPTOGRAPHY MODELS
Author:	DR. C.NAGA GANESH, DR. MAMATHA G , DR. VIVEK VEERAIAH , DR. DEVIKA RANI ROY , AMARJA ADGAONKAR , MRS. SUPRIYA SANJAY AJAGEKAR , ANKUR GUPTA, DR. PATIL AMIT MADHUKAR
Abstract:	5G networks, IoT devices, and cloud computing have all made it easier for people to come up with new ideas, but they have also made systems more vulnerable to increasingly complicated cyber assaults. Traditional security measures work to some extent, but they can't handle the complicated, limited resources, and constantly changing digital environments we live in today. This paper introduces an innovative framework for creating resilient, scalable, and flexible security systems via the integration of artificial intelligence and advanced cryptography methodologies. A hybrid deep learning model that incorporates CNN, LSTM, and Autoencoders (AE) is utilised to detect threats. This model is more accurate than baseline methods and has fewer false positives. Dynamic key rotation across IoT, mobile, and cloud infrastructures may improve cryptographic resilience via reinforcement learning-driven key management. Blockchain-based trust management makes transactions and contracts more accountable and open by lowering communication costs and making sure that model training that protects privacy is done via federated learning. Experimental evaluations of healthcare, IoT, and 5G datasets indicate that the system exhibits enhanced accuracy (94.3%), increased resilience to adversarial attacks (with an attack success rate reduction of 70%), and improved energy efficiency. Multi-objective optimisation has shown that the framework's performance, scalability, and security balance is good. This makes it a good choice for next-generation digital infrastructures. This work aims to include post-quantum solutions and explainable AI into future advancements, facilitating the convergence of AI and cryptography to enhance security.
Keywords:	AI, Cryptography, Federated Learning, Blockchain, Reinforcement Learning, Deep Learning, Cybersecurity, IoT, 5G Networks, Post-Quantum Cryptography, Trust Management, Secure Data Management.
DOI:	https://doi.org/10.5281/zenodo.18824315
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	HYBRID DEEP LEARNING AND OPTIMIZATION-DRIVEN FRAMEWORK FOR ENHANCED SKIN CANCER DETECTION AND CLASSIFICATION
Author:	A CHANDRA SEKHAR , D VENKATESH
Abstract:	Globally, skin cancer is one of the most common diseases caused by excessive exposure to ultraviolet (UV) radiation. The outcomes and prognosis for treating skin malignancies significantly improve with early detection and accurate diagnosis. However, many diagnostic methods have limitations, entail high computational costs, rely heavily on manual feature extraction, lack good generalization across datasets, and are susceptible to adversarial attacks. This paper addresses these limitations in skin cancer diagnosis by proposing an improved Wavelet-AHE Diffusion Enhanced Hybrid Network (WADE-HNet) for enhancing detection as well as classification. The proposed ensemble technique integrates ResEff-FuseNet for feature extraction, Firefly-Bitterling Adaptive Selection Optimization (FBASO) for optimal feature selection, Multi-Stage Attention Capsule Network (MSA-CapsNet) for enhanced classification performance, and Modified U-Net++ for lesion segmentation. Such improvements increase the interpretability, generalizability, and computational efficiency of model performance. Finally, empirical results for both the HAM10000 and ISIC 2019 datasets validate that WADE-HNet outperforms benchmark models with a high accuracy of 99.39%. In this way, the proposed strategy ensures consistency in clinical usage while reducing false positives and false negatives.
Keywords:	Skin Cancer Detection, DL, Feature Selection, Capsule Networks, Optimization, Image Processing, WADE-HNet
DOI:	https://doi.org/10.5281/zenodo.18824335
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	EFFICIENT COFFEE PLANT DISEASE PREDICTION WITH ADAPTIVE FEATURE SELECTION AND DEEP CONVOLUTIONAL NEURAL NETWORK
Author:	P.GOBINATH ,Dr.M.RAMASWAMI
Abstract:	The problem of predicting disease in coffee plant is briefly analyzed and there exist numerous techniques around the issue. However, the approaches struggle to achieve maximum accuracy in predicting possible disease which affects the growth of plant. Towards this, an efficient Adaptive Invariant Feature Selection and Approximation based Disease Prediction with DCNN (AIFSADP-DCNN) is sketched. The method applies Intensity Deviation Normalization technique to normalize the leaf image and enhance the image quality Further, Color Quantization Segmentation algorithm is enforced towards segmenting the coffee plant features. Next, color, texture, and distribution features are extracted from the segmented image. Concern features extracted are trained with deep convolution neural network with three layers of convolution and pooling layers. The output layer neurons are designed to measure Color Constraint Support (CCS), Texture Constraint support (TCS) and Distribution Constraint Support (DCS) towards various classes of features. Finally, the method estimates plant disease support (PDS) towards various classes based on which the method identifies the disease class. The proposed model improves the performance of disease prediction in coffee plants with higher accuracy up to 96.5%.
Keywords:	Disease Prediction, Plant Management, Deep Learning, CNN, AIFSADP-DCNN, PCS, DCS, TCS, CCS
DOI:	https://doi.org/10.5281/zenodo.18824350
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	BIG DATA AS A DECISION-MAKING TOOL IN HUMAN RESOURCE MANAGEMENT
Author:	DMYTRO KOBETS , IRYNA CHERNOVA , TARAS MUKHA , VUGAR SALMANOV , SVITLANA ALEKSANDROVA
Abstract:	The relevance of the study is determined by the need for high-performance data processing architectures to increase the efficiency of human resource management (HR) in the context of digital transformation and the increasing complexity of analytical processes. The aim of the research is to create an optimized data-driven human resource (HR) decision-making architecture with the integration of machine learning (ML) modules, adaptive pipelines, and streaming analytics, verified by Unified Modelling Language (UML) and metric comparison. Research methods: SWOT analysis, metric comparison, structural optimization decomposition, UML, comparative analysis of UML models, metric comparison of optimized architecture. Optimized HR Data-Driven Decision-Making architecture provided an increase in predictive accuracy to 0.962 (+8.4%), metric stability (+9.1%), and discriminability (+7.8%). SWOT analysis identified 6 strengths, 5 critical weaknesses, 4 risks; comparison confirmed the superiority of SAP Analytics Cloud (0.94; 0.91; 0.89) over the market average (0.81; 0.78; 0.75). UML integration of adaptive feature engineering pipelines, cognitively optimized ML modules, and streaming analytics increased interoperability (+14.6%), modular integrity (+12.3%) and algorithmic resilience (+15.1%). The academic novelty of the study is the formalization and verification of an optimized HR Data-Driven Decision-Making architecture that integrates ML modules, feature engineering, and stream analytics, providing a predictive accuracy of 0.962 (+8.4%) and interoperability of +14.6%. Prospects for further research include localized implementation of the architecture in a test environment with metric and cognitive functional verification, iterative optimization of modules based on the results of institutional contextual empirical testing.
Keywords:	Data Analytics, Human Resource Management, System Architecture Modelling, Decision-Making Architecture, Data-Driven Analytics, Predictive Modelling, Organizational
DOI:	https://doi.org/10.5281/zenodo.18824366
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	DEVELOPING A HYBRID LOG-FILE–BASED COVERT CHANNEL FOR SECURE DIGITAL FORENSIC DOCUMENT TRANSMISSION
Author:	TAMER S. FATAYER, SAMY S. ABU-NASER , MUSBAH J. MOUSA
Abstract:	Security is a fundamental requirement in Digital Forensic (DF) systems, given the increasing number and sophistication of cyber threats. Among the major challenges in DF is ensuring the secure transmission of Digital Forensic Documents (DFDs) between authorized entities. This paper proposes a hybrid covert channel designed to enhance the confidentiality and un-detectability of DFD transfers by exploiting runtime errors and system log files. The proposed method leverages the Digital Forensic Legal Affairs Server (DFLAS) log file, where original DFDs embedded with covert data are written and subsequently transmitted through a concealed communication channel. The original files remain preserved in the server’s log file after transmission, ensuring both integrity and traceability. To achieve robust security, the hybrid covert channel integrates randomization, encryption, and controlled packet delay, collectively reinforcing confidentiality and resistance to detection. The performance evaluation focuses on analyzing time transmission and bandwidth efficiency during exchanging the DFDs. Experimental results show that the proposed channel achieves high efficiency in covert data transmission, with transmission time and throughput comparable to conventional channels. Additionally, it provides a high level of confidentiality through the use of encryption and randomization techniques. The results show that transferring a 112 MB file requires approximately 4000 ms on the DFLAS side, demonstrating competitive performance compared with existing approaches. The results further indicate that the extraction capacity rate varies according to the applied delay intervals while maintaining practical throughput levels suitable for DF real-world deployment. The achieved throughput of 29.12 MB/sec confirms the method’s practicality. The proposed covert channel provides confidentiality, plausibility, and undetectability, making it a promising approach for secure and efficient transmission of digital forensic documents.
Keywords:	Covert channel, Digital forensics, Log file, Undetectability, Encryption.
DOI:	https://doi.org/10.5281/zenodo.18824392
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	AI-READY VR/360° E-LEARNING PLATFORM FOR DISTANCE EDUCATION: SYSTEM ARCHITECTURE, DEPLOYMENT, AND ADOPTION EVIDENCE
Author:	ANASS TOUIMA , MOHAMED MOUGHIT
Abstract:	The sudden move to distance learning during the COVID-19 pandemic forced universities to rely heavily on videoconferencing tools and basic learning management systems. While these solutions kept courses running, many students reported fatigue, disengagement, and difficulties visualising complex concepts. In response to these challenges, this study describes the design and preliminary evaluation of a Virtual Reality (VR) e-learning platform, extended with an AI-ready facial expression recognition module and a 360° video web alternative, deployed in a Moroccan higher-education context. The platform allows students to access an immersive virtual campus and interactive learning scenarios either through VR headsets or through 360° video on standard devices. A total of 106 students from the Higher Institute of Information and Communication (ISIC) interacted with the system and then completed a structured questionnaire about their experience, perceived usefulness, and intention to adopt the platform. Descriptive analyses were complemented with chi-square and Spearman correlation tests to explore differences between academic levels and the role of prior exposure to immersive media. The results show high awareness of VR and 360° video, strong perceived added value of VR for distance learning, and broad support for regular use of the platform, with no major differences between Licence and Master students and only weak associations with previous VR experience. These findings position the proposed AI-ready VR environment as a promising complement to conventional distance learning tools and motivate future controlled studies focusing on objective learning outcomes and the full integration of AI-driven emotion tracking.
Keywords:	Distance Learning, Virtual Reality, VR, Artificial Intelligence, Higher Education
DOI:	https://doi.org/10.5281/zenodo.18824411
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	METAVERSE-BASED VIRTUAL EDUCATION PLATFORMS USING BLOCKCHAIN FOR CREDENTIAL VERIFICATION
Author:	DR. VIVEK VEERAIAH , DR. M.YELLAIAH NAYUDU , DR. MAMATHA G , DR. TARUN DALAL , M AHMER USMANI , DR. VIPIN JAIN , MR. VINOD MOTIRAM RATHOD , DR. HUMA KHAN
Abstract:	Digital classrooms now include a lot of safe methods to preserve and exchange information, as well as engaging technology. This article suggests a Metaverse virtual school network based on blockchain technology as a way to make sure that credentials are real and safe. However, much study on blockchain-based credentialing systems or metaverse learning environments is theoretical or survey-based. Credential reliability, integration with current systems, scalability, and user confidence in metaverse learning platforms are little researched. Studies with an integrated, implementation-oriented approach and empirical evaluation are beneficial. This research introduces and evaluates a Metaverse Virtual Education Platform that uses blockchain technology to validate credentials transparently and securely. Our solution solves the issues that conventional learning management systems (LMS) have with being transparent and honest about student information, keeping credentials safe, and persuading people to utilize them. While there is increasing interest in both blockchain-based credentials and metaverse-based education, existing solutions currently lack a cohesive and validated methodology. Secure, dependable, and easily verifiable academic credentials are paramount within decentralised virtual learning environments; this system is indispensable for realising these objectives. The platform's effectiveness has been demonstrated through multiple trials. The metaverse platform, when compared to its competitors, exhibits a higher degree of consumer engagement, evidenced by an average usage duration of 41 minutes, in contrast to the 22 minutes observed on alternative platforms, and a superior satisfaction rating of 4.7 out of 5. Although blockchain effectively mitigates the risk of identity theft and fraud, it does introduce a slight delay in the credential validation process; Hyperledger requires 2.1 seconds, while Ethereum necessitates 1.8 seconds. Furthermore, a security analysis indicates that unauthorised modifications are more readily identifiable, and MFA is streamlined. The system's responsiveness is unaffected by concurrent user load. Navigation, design, effectiveness, credential trust, and user satisfaction constitute five key usability metrics, and the system demonstrates strong performance in each area. This research suggests that metaverse designs incorporating blockchain-based certification systems can facilitate the development of secure, engaging, and novel online learning environments.
Keywords:	Metaverse, Virtual Education, Blockchain, Credential Verification, Digital Credentials, Decentralized Systems, Education Security, Trust Management
DOI:	https://doi.org/10.5281/zenodo.18824426
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	AN INTERPRETABLE DEEP LEARNING BASED DECISION SUPPORT SYSTEM FOR STOCK FORECASTING AND INVESTMENT RECOMMENDATION USING TEMPORAL FUSION TRANSFORMER
Author:	D. MANJU , M. KOTESWARA RAO , RAJESH KUM AR VERMA , PADMINI DEBBARMA , ANAND KUMAR SARASWATHI RATHOD
Abstract:	The stock market has rapid changes caused by economic conditions, company performance and investor behavior. Because of this, stock price predictions and correct investment decisions are hardly achieved. Investors today use computer-based systems to study large volumes of financial data and find useful patterns barely perceived using traditional methods. However, most existing stock prediction and recommendation systems that have some major problems. Many systems use only technical indicators or only fundamental data, while not both together. Some models cannot properly capture both the short-term and long-term price movements of the stock. Moreover, many of these deep learning models behave like black boxes and fail to explain clearly why any stock is recommended to be on buy, sell or hold. This therefore compromises trust in their prediction capabilities, especially those specifically for highly volatile markets. In this paper, we try to solve those problems by proposing a new stock forecasting and recommendation system based on the Temporal Fusion Transformer (TFT), combining historical stock prices, technical indicators and fundamental financial data in one framework. It can learn both short-term and long-term trends and show which factor is most important for the prediction. After predicting the price and other financial indicators, our model extends TFT into a decision-support system by embedding financial ratio analysis and recommendation logic within the forecasting pipeline.
Keywords:	Deep Learning, Investment Recommendation, Stock Forecasting Temporal Fusion Transformer, Technical Indicators, Fundamental Analysis.
DOI:	https://doi.org/10.5281/zenodo.18824437
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	USE OF ARTIFICIAL INTELLIGENCE MODELS IN EDUCATIONAL DATA MANAGEMENT TO SUPPORT PEDAGOGICAL DECISIONS
Author:	OKSANA KARABIN , NATALIYA LUPAK , VUGAR IBRAGIM OGLY SALMANOV , SERHII KRYVOSHLYKOV, ZINOVII ONYSHKIV
Abstract:	The relevance of the topic is determined by the growing role of analytical systems in modern education and the need to combine forecasting accuracy with the efficiency and stability of algorithms. The paper compares three configurations – ML-Base, DL-Advanced and Hybrid-Assist – using the integral educational analytics effectiveness index (IEAEI), which combines indicators of accuracy (Acc), stability (Stab), processing time (Time) and interpretability (Interpret). The methodology included normalisation of results, non-parametric hypothesis testing (Kruskal–Wallis and Mann–Whitney criteria), analysis of changes in weight coefficients (ΔW-index) and assessment of correlations between platform dynamics and forecast stability (Spearman's coefficient). To test the transferability of the findings, a design with three contexts – Ukraine, Azerbaijan and Poland – was used, which made it possible to compare the behaviour of models in countries with transition economies and in European educational environments. The results showed that the hybrid configuration provides the best balance between stability and accuracy (IEOA > 0.84; cross-country: Ukraine – 0.821; Azerbaijan – 0.818; Poland – 0.824) with a relatively fast adaptation time; DL-Advanced achieves higher maximum accuracy, especially under conditions of more complete and consistent data (approaching Hybrid-Assist in the Polish subsample), but requires more time for convergence; ML-Base has the shortest response time in all three contexts, but is inferior in terms of predictive quality. The scientific novelty lies in the generalisation of the patterns of interaction between the algorithm type, platform characteristics and pedagogical practice requirements in an international context, which allows for the justified integration of intelligent models into educational analytics. Prospects for further research include expanding the set of evaluation metrics with indicators of cognitive convenience for users and adapting the methodology to cloud and mobile educational solutions, taking into account cross-country data variability.
Keywords:	Pedagogy, Knowledge, Learning, Educational Analytics, Ml-Base, Dl-Advanced, Hybrid-Assist, Ieoa, Δw-Index, Forecast Stability, Pedagogical Solutions, Non-Parametric Tests
DOI:	https://doi.org/10.5281/zenodo.18824453
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text

Title:	A COMPARATIVE STUDY OF MACHINE LEARNING FOR AGE GROUP IMPUTATION IN LARGE-SCALE E-COMMERCE DATA
Author:	JURON PAIK
Abstract:	Current privacy regulations and the prevalence of voluntary non-disclosure have led to significant gaps in demographic data within e-commerce platforms, severely hindering personalized marketing efforts. This study proposes a machine learning–based approach to address the problem of missing age group information in large-scale e-commerce platforms. Resolving this issue can enhance the personalization potential while respecting user privacy constraints. The dataset used in this research comprised actual e-commerce platform data, including customer behavior logs, product attributes, and temporal and regional variables. Initially, five classification models—logistic regression, decision tree, random forest, k-nearest neighbors (knn), and XGBoost—were compared. Preliminary experiments revealed that logistic regression performed relatively poorly; therefore, it was excluded from the final comparison, and hyperparameter optimization was performed on the remaining four models. Model performance was evaluated on the validation set using accuracy and F1-score as the primary metrics. Experimental results showed that the random forest (default configuration) achieved the highest performance with approximately 78\% accuracy, while XGBoost, although underperforming in the default setting, improved to a comparable level after optimization. In contrast, decision tree and knn showed limited improvement from optimization, with performance in some cases declining compared to the default setting. Feature importance analysis identified behavioral frequency, a specific event type, gender, and product attributes as key factors. This research contributes by empirically demonstrating the feasibility of constructing age group prediction models using large-scale e-commerce data, thereby offering a practical strategy for addressing missing demographic information under privacy constraints. Furthermore, the feature importance analysis provides actionable insights for target marketing and personalized recommendation system design. Conclusively, this study empirically demonstrates that behavioral logs alone are sufficient to predict demographic attributes with high accuracy. The proposed Random Forest-based framework offers a cost-effective and privacy-preserving alternative to complex deep learning models for practical deployment in real-world e-commerce systems.
Keywords:	Age Group Prediction, Machine Learning, Privacy-Preserving Data Analysis, Personalized Recommendation, E-commerce Data
DOI:	https://doi.org/10.5281/zenodo.18824465
Source:	Journal of Theoretical and Applied Information Technology 28^th February 2026 -- Vol. 104. No. 4-- 2026
Full Text