		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
March 2026 | Vol. 104 No.5

Title:	DOLLMAKER GIANT ARMADILLO OPTIMIZATION ENABLED VISION TRANSFORMER CONVOLUTIONAL FORWARD HARMONIC NET FOR STRESS DETECTION USING PPG SIGNAL
Author:	C.S.L VIJAYA DURGA, J.MANIMARAN, M PURUSHOTHAM REDDY
Abstract:	Stress detection is crucial because early identification allows for timely intervention that prevents the development of serious health problems related to prolonged stress. However, existing methods suffer from limited feature representation, ineffective feature fusion, poor robustness to noise, class imbalance issues, and suboptimal optimization during model training. Many conventional machine learning (ML) and deep learning (DL) approaches fail to capture the complex and nonlinear physiological patterns associated with stress, leading to reduced generalization and reliability. To address these challenges, this work proposes a Dollmaker Giant Armadillo Optimization-enabled Vision Transformer Convolutional Forward Harmonic Net (DGAO_ViTCFHNet) model for detecting stress using Photoplethysmogram (PPG) signals. Initially, the input PPG signal is passed into a feature extraction process, and the outcome is considered as output-1. Moreover, the same input PPG signal undergoes a feature extraction process, including time-domain-based features, statistical features, and frequency-based features, and the outcome is considered as output-2. These extracted features are then fused using the Squared-chord distance with Quantum Dilated Convolutional Neural Networks (QDCNN). The fused features are then subjected to data augmentation, which is accomplished by the oversampling technique. At last, detection is performed using ViTCFHNet, which integrates Vision Transformer (ViT) and Convolutional Neural Networks (CNN) with a forward harmonic analysis concept. ViTCFHNet is trained using Dollmaker Giant Armadillo Optimization (DGAO), which is derived by integrating Dollmaker Optimization Algorithm (DOA) and Giant Armadillo Optimization (GAO). The effectiveness of DGAO_ViTCFHNet is analyzed by assuming the metrics, like accuracy, Positive Predictive Value (PPV), Negative Predictive Value (NPV), True Positive Rate (TPR), and True Negative Rate (TNR), that obtained superior values of 96.05%, 87.88%, 95.74%, 88.05%, and 87.80%.
Keywords:	Quantum Dilated Convolutional Neural Networks, Vision Transformer, Convolutional Neural Networks, Dollmaker Optimization Algorithm, Giant Armadillo Optimization.
DOI:	https://doi.org/10.5281/zenodo.19127908
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	THREATSCAN - CYBERTHREAT PREDICTION SYSTEM USING TEMPORAL CONVOLUTIONAL NETWORKS
Author:	THONDEPU ADILAKSHMI, MALLANNAGARI SUNITHA, MADDARAPU PAVAN DURGA NIVAS, HARSHITHA PALLAPOLU
Abstract:	The growing digitization of the public and private sectors increases attack surfaces for cyber threats including cyber denial-of-service (DoS) attacks and more advanced zero-day exploits. Unlike established Intrusion Detection Systems (IDSs) that utilize rule-based signatures and conventional machine learning, the proposed AI-based systems which utilize Temporal Convolutional Networks (TCNs) anticipates threats by scrutinizing multivariate time-series network traffic and system log data of an organization. The proposed model leverages the Cybersecurity Threat Detection and Awareness Program dataset which includes real-world network event data associated with flow logs, IDS alerts, firewall telemetry, and threat intelligence feeds from India. The model evaluates threats and anticipates cyber-attack vectors within the scope of multi-class classification and anomaly detection. The choice of TCNs is due to their ability to model long-range temporal features that is essential in recognizing and predicting attacks that evolve over time. The model predicts and classifies attack types, and signals a detection of an anomaly using five levels (No Threat, Low, Medium, High, and Critical). TCN provides security analytics and threat intelligence convergence which is instrumental to achieving a higher efficacy of incident response, real-time anomaly detection and risk scoring help security teams prioritize responses and mitigate risks before threats escalate.
Keywords:	Cyberthreat, Intrusion Detection System, Dual Stacked Temporal Convolutional Network, Temporal Convolution Network, Machine Learning.
DOI:	https://doi.org/10.5281/zenodo.19128006
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	DESIGNING FOR USABILITY: AN AI-HUMAN HYBRID MENTAL HEALTH COUNSELING WEB APP FOR STUDENTS
Author:	ELGA THERESIA, TANTY OKTAVIA
Abstract:	The significant rise in demand for mental health support among students has strained counselors and psychologists’ capacity to serve, necessitating innovative approaches that extend access for the students while reducing professional workload. AI is an increasingly explored technology to solve this issue. This research investigates and formulates the user experience of an AI-Human hybrid mental health counseling web app for students. A mixed-methods design was adopted. It consisted of in-depth interviews with students and mental health professionals to elicit user-centered insights, and an online survey using the Kano model and conjoint analysis to prioritize system features. Findings consistently highlight support for a hybrid human–AI model. Both students in higher education and mental health professionals recognized AI’s potential as a complementary tool, particularly in reducing administrative tasks. Participants favored the use of AI for onboarding clients and collecting initial information, thereby enabling counselors to dedicate more time to meaningful dialogues. At the end of sessions, AI was also viewed as valuable in generating summaries and supporting post-session analysis. However, attitudes towards AI use during live counseling were more cautious. This research contributes conceptual and empirical foundations for the development of an AI-enhanced solution that preserves human connection while improving efficiency and quality of care.
Keywords:	Artificial Intelligence, Mental Health, User-Centered Design, Counseling, Human-Computer Interaction
DOI:	https://doi.org/10.5281/zenodo.19128056
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	REVOLUTIONIZING WASTE CLASSIFICATION THROUGH MULTIMODAL CONDITIONAL GANS WITH ROBUST REGULARIZATION
Author:	NANDIKANTE SHRAVANI, Dr. S. VIGNESHWARI
Abstract:	This study proposed a novel framework for automated waste classification by employing an enhanced Conditional Generative Adversarial Network (cGAN) integrated with multimodal data fusion using thermal and hyperspectral imagery. To address the persistent challenge of adversarial training instability, spectral normalization and gradient penalty regularization were incorporated, which ensured stable convergence and robust feature learning. The proposed framework not only generated high-quality and diverse synthetic waste samples but also resulted in measurable improvements in downstream classification performance. Experimental evaluations conducted on public benchmark datasets demonstrated an approximate 15% improvement in classification accuracy and a 20% enhancement in robustness compared to conventional approaches. The results highlighted the effectiveness of combining multimodal sensing with advanced generative modeling for improving discrimination capability under data-scarce and variable conditions. Overall, the findings established the proposed approach as a scalable and reliable solution for accurate waste categorization, underscoring the potential of multimodal machine learning frameworks in supporting environmentally sustainable waste management systems.
Keywords:	Conditional Generative Adversarial Networks; Waste Classification; Multimodal Data; Regularization Techniques; Spectral Normalization; Gradient Penalties; Environmental Sustainability; Machine Learning; Hyperspectral Imaging; Thermal Imaging
DOI:	https://doi.org/10.5281/zenodo.19128125
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	INTEGRATING ADVERSARIAL AUTOENCODERS WITH GATED RECURRENT UNITS TO IDENTIFY ANOMALIES IN ECG SIGNALS
Author:	SHAIK JANBHASHA, VENKATESWARLU SUNKARI, DIPANWITA DEBNATH, VENKAT RAO PASUPULETI, AKETI NARESH, LAKSHMANARAO TALAPAKULA, JUTU GOPAIAH, ROHTIH BALA JASWANTH B
Abstract:	Cardiovascular diseases remain a leading global health concern, necessitating advanced methods for early detection of cardiac anomalies. This study proposes a hybrid model combining Adversarial Autoencoders (AAE) and Gated Recurrent Units (GRU) to improve anomaly detection in Electrocardiogram (ECG) signals. The AAE module extracts robust latent representations of normal ECG patterns, while the GRU captures temporal dependencies within the signal. Experimental results demonstrate superior performance compared to existing methods, achieving 98.6% accuracy, 97.7% recall, 98.2% F1-score, 98.8% precision, and 99.1% AUC-ROC. The proposed framework reduces false positives and enhances diagnostic reliability, offering a promising tool for automated cardiac monitoring.
Keywords:	ECG, Anomaly Detection, Adversarial Autoencoder, GRU, Deep Learning
DOI:	https://doi.org/10.5281/zenodo.19128170
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	COGNITIVE ALLOCATOR: AN END-TO-END SECURE AND DYNAMIC VIRTUAL MACHINE ALLOCATION FRAMEWORK FOR OPTIMIZED RESOURCE MANAGEMENT IN CLOUD DATA CENTERS
Author:	K SARAVANAN, DR.R.SANTHOSH
Abstract:	Cloud Computing (CC) offers a ubiquitous service to the Information Technology (IT) environment with higher resources. However, due to massive usage of cloud resources the availability of services often faced with service outage, power consumption, and security issues. Several state-of-the-art works tends to tradeoff the resource usage optimization and security in cloud Virtual Machines (VMs). To overcome this issue, we design a secure and effective cloud VM allocator named “Cognitive Allocator”. The designed Cognitive Allocator model ensures trade-off among the end-to-end security and resource management in the cloud computing environment by utilizing Deep Learning (DL), Deep Reinforcement Learning (DRL), and Optimization algorithms respectively. The entities involved in the model includes Cloud Users (CUs), Cloud Service Providers (CSPs), Authentication Server (AS), Edge Server (ES), and Cloud Broker. Firstly, the CUs and CSPs are authenticated to the AS for guaranteeing the authenticity for jeopardizing the network traffic attacks. The authenticated CUs are then restricted by the ESs with its hybrid access control (Role and policy-based access control) mechanism using Dual Agent DRL (DA-DRL) algorithm. The DA-DRL algorithm learns the past experience and firmly controls access for the CUs based on their roles, Service Level Agreement (SLA), and their authenticity. Finally, we perform secure and optimized VM allocation in the cloud broker server using DL and optimization algorithm named Attention Classification Network (AC-Net) and Horse Herd Optimization (H2O) respectively. The AC-Net classifies the user task into three classes such as public, confidentiality, and highly sensitive. Based on the classified user task and security parameters, the H2O algorithm effectively allocates the VM for the CSPs. The proposed work is implemented using Cloud Sim of version 3.03 with various validation metrics such as security and privacy analysis, cost execution, resource utilization, power consumption, and allocation time. From the validation results, the proposed cognitive allocator model outpaces than the state-of-the-art models.
Keywords:	Cloud Computing (CC), Virtual Machine (VM), Cloud Service Providers (CSPs), Edge Server (ES), Deep Learning (DL), and Deep Reinforcement Learning (DRL)
DOI:	https://doi.org/10.5281/zenodo.19128224
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	EFFECTIVE CBIR BASED ON HYBRID IMAGE FEATURES AND MULTILEVEL APPROACH
Author:	V. ARCHANA REDDY, V. VIJAYA KUMAR
Abstract:	The instantaneous search and retrieval of the most relevant images to a specific query is one of the significant applications of image processing. The process of retrieval of images using image contents is widely known as content base image retrieval (CBIR). The image features extracted from the local windows of 33 resulted good results in CBIR. However, the micro windows of 22 derived the texton and motif features and played a dominant role in CBIR. This paper segmented the 33 window into two windows namely cross and diagonal windows. Four directional motifs of 13 are extracted from each of the cross and diagonal segments. The motif features are derived using Rule based Directional motif (RDM) to address the ambiguity issues. This paper transformed the motif indexes derived on a 1x3 triangular window to a 33 window and derived eight triangular RDM local units. The local features are extracted by integrating the features extracted from the eight different images. The extracted features hold directional, textures, patterns, edge properties derived from the 33 and triangular windows. The results indicate the superiority of the proposed methods.
Keywords:	33 Window, 13 Triangular Window, Directional Motif;
DOI:	https://doi.org/10.5281/zenodo.19128259
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	HNBK-MR-BILSTM HYBRID MACHINE LEARNING MODEL FOR ENHANCED DIABETIC RETINOPATHY DETECTION IN FUNDUS IMAGING
Author:	ALKA SINGH, RAKESH KUMAR, AMIR H GANDOMI
Abstract:	Utilizing a Computer-Aided Diagnosis (CAD) system of retinal fundus images is becoming optimal for manual fundus inspection. In addition, CAD of retinal fundus images was found to be more reliable and necessitated less time for both processing and analysis. Although several preprocessing techniques have been proposed, preprocessing remains challenging owing to a high level of contrast in the retinal vasculature network and image quality. In this work, a method called Hessian Niblack’s Binarization Keypoint and Multi-scale Retinex Bidirectional LSTM (HNBK-MR-BiLSTM) for Diabetic Retinopathy detection is proposed. The HNBK-MR-BiLSTM is split into three sections: preprocessing, feature extraction, and classification for diabetic retinopathy detection. Based on the notion that the vessel profile could be modeled by separating into three channels, red, green and blue, the Hessian Frangi filter tries to model and preprocess the input images by retaining only the green channel while eliminating the red and blue channels for better representation of different types in fundus images via Second-order Partial Derivatives. Second, ophthalmoscopy was removed and combined the features from retina images based on Niblack’s Threshold by binarization, as well as the coefficients of the rectangular array of pixels via PixMap. Finally, Bidirectional Multi-scale Retinex LSTM is applied to the keypoints detected for accurate classification between five distinct classes of images. Experimental results show that the proposed HNBK-MR-BiLSTM method for diabetic retinopathy learns more features of small targets and can efficiently enhance the classification performance of diabetic retina. The performance analysis of the proposed method to achieve better accuracy, time, sensitivity, specificity, and Peak Signal Noise Ratio (PSNR) compared to conventional methods is described. The results show that the proposed HNBK-MR-BiLSTM method improves at diagnosing diabetic retinopathy than more existing methods.
Keywords:	Diabetic Retinopathy, Hessian Frangi, Second-order Partial Derivatives, Niblack’s Binarization, Keypoint, Nonlinear Feature Extraction, Multi-scale Retinex, BiLSTM
DOI:	https://doi.org/10.5281/zenodo.19128318
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	ANALYZING LAGGED CAUSALITY OF MULTIPLE MARKET FACTORS ON BITCOIN PRICE: AN INTERDISCIPLINARY APPROACH
Author:	SKY NURIMBA, TANTY OKTAVIA
Abstract:	The cryptocurrency market, particularly Bitcoin, is characterized by high volatility influenced by various factors including public sentiment on social media, trading volume, and psychological market indicators such as the Fear and Greed Index (FGI). This study analyzes the lagged relationships and directional causality between Platform X sentiment, Bitcoin trading volume, the FGI, and Bitcoin price movements through an interdisciplinary approach that integrates Natural Language Processing and econometric modeling. The novelty of this research lies in its comprehensive multivariate framework that simultaneously examines bidirectional causality among four key market variables using optimal lag selection (10 periods) and the Toda-Yamamoto procedure, addressing critical gaps in temporal dynamics that previous studies overlooked. Sentiment data from Platform X was extracted using FinBERT, a model optimized for financial text analysis, and Vector Autoregression together with Granger Causality Tests were implemented to identify temporal lag patterns and causal relationships over the period from March 2022 to March 2023. The results indicate that Bitcoin price exerts a significant influence on Platform X sentiment, Bitcoin trading volume, and the FGI, with trading volume being the only variable that also exhibits a reverse influence on Bitcoin price. These findings suggest that price movements play a dominant role in shaping sentiment, trading activity, and market psychology, while trading volume maintains a feedback loop with price. This study provides both practical and theoretical contributions by offering a comprehensive framework for understanding the dynamics of the cryptocurrency market and the interactions among social media sentiment, trading behavior, and investor psychology in relation to Bitcoin price movements.
Keywords:	Machine Learning, Sentiment Analysis, FinBERT, Cryptocurrency, Granger Causality
DOI:	https://doi.org/10.5281/zenodo.19128350
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	AVAILABILITY AND IMPACT OF ARTIFICIAL INTELLIGENCE SYSTEMS ON DECISION MAKING SUPPORT IN U.S. SMALL AND MEDIUM-SIZED BUSINESSES
Author:	KATERYNA HALAN
Abstract:	Artificial intelligence (AI) systems are becoming a critical tool for digital transformation, but their availability to small and medium-sized businesses (SMBs) remains limited and uneven. High cost, lack of human resources (HR), and low digital maturity create barriers that hinder the implementation of innovations in enterprises in this segment. The aim of the research is to assess how the availability of AI-based decision support systems (DSS) affects the likelihood of their implementation in the United States (USA). The study also determines how such technologies change labour productivity, operating costs, forecast accuracy, and the speed of management decisions in SMBs. The methodology is based on panel data from 2022–2024 from The Business Trends and Outlook Survey (BTOS), Annual Business Survey (ABS), and the Federal Reserve Small Business Credit Survey. An AI Availability Index (AIAI) is proposed, created using the principal components method, which combines indicators of cost, digital maturity, integration, HR, and telecommunications infrastructure. The results show that availability significantly increases the likelihood of AI-DSS implementation in SMBs in the USA. Empirical estimates show an increase in labour productivity by about five percent and a decrease in the share of operating costs by more than three percentage points. The accuracy of forecasts improves, and the speed of management decisions increases because of the integration of AI systems. The effects are significantly stronger for high AIAI values, confirming the key role of infrastructure and digital maturity. Practical conclusions include the development of quick-start roadmaps for SMBs and the implementation of risk management protocols under the NIST AI Risk Management Framework (RMF). The academic novelty is the creation of a unique AIAI and conducting a causal analysis of the effects on key performance indicators (KPIs), taking into account heterogeneity.
Keywords:	Artificial Intelligence, Decision Support Systems, Small and Medium-Sized Business, Availability, Digital Transformation, Productivity, Econometric Modelling
DOI:	https://doi.org/10.5281/zenodo.19128399
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	IMPROVEMENT OF ANFIS FOR EARLY CARDIOVASCULAR DISEASE PREDICTION MODEL USING DIFFERENTIAL EVOLUTION (DE) AND LIME EXPLAINABLE AI
Author:	SRI SUMARLINDA, WIJI LESTARI, FAULINDA ELY NASTITI
Abstract:	The early detection of cardiovascular disease (CVD) remains a critical challenge in preventive healthcare, requiring predictive models that are both accurate and interpretable. This study aims to develop an early CVD prediction model enhanced with Differential Evolution (DE) optimisation and explainable AI using LIME. The models were evaluated on a dataset of 500 samples comprising six features: age, body mass index (BMI), systolic blood pressure, diastolic blood pressure, cholesterol, and blood sugar. The ANFIS model was improved by optimising its premise and consequent parameters through mutation, crossover, and selection processes within the DE algorithm. LIME was employed to provide interpretability by revealing the contribution of each feature to the ANFIS-DE prediction outcomes. The dataset was evaluated using three different data-splitting schemes to identify the most effective training testing proportion. In Model 1, the data were divided into 60% for training and 40% for testing. Model 2 applied a 70%:30% split, whereas Model 3 used an 80%:20% split. Among these configurations, Model 3 (80%:20%) yielded the highest predictive performance, indicating that a larger portion of training data contributed to better model generalisation and overall accuracy. The enhanced ANFIS-DE model outperformed the baseline ANFIS, achieving higher testing accuracy (0.9200 vs. 0.9167), precision (0.9290 vs. 0.9287), recall (0.9250 vs. 0.9250), and F1-score (0.9425 vs. 0.9367), alongside a lower error value (0.2011 vs. 0.2019). LIME analysis further indicated that blood sugar had the highest contribution (0.27), followed by systolic blood pressure (0.25), age (0.20), cholesterol (0.06), BMI (0.03), while diastolic blood pressure exerted a slight negative influence (–0.01), demonstrating the usefulness of feature-level explanations in supporting early CVD risk prediction.
Keywords:	Cardiovascular Disease, ANFIS, Differential Evolution (DE), LIME Explainable AI, Early Prediction Model
DOI:	https://doi.org/10.5281/zenodo.19128449
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	PECULIARITIES OF THE USE OF BLOCKCHAIN TECHNOLOGIES IN DOCUMENTING AND AUTHENTICATION OF EVIDENCE DURING CYBER INVESTIGATIONS
Author:	VIACHESLAV KULIUSH, VLADYSLAV VEKLYCH, NATALIIA IAKYMCHUK, SERGIY MARCHEVSKYI, IVAN KURYLIN
Abstract:	As cybercrimes rise, effective methods to preserve and evaluate digital evidence are needed. This study examines how blockchain technology might enhance cyber investigation, digital evidence capture, and verification. This investigation compares public and permissioned blockchain systems' legal and forensic reliability. Legal and forensic comparisons were conducted in Estonia, Germany, Ukraine, and the Netherlands. The inquiry focused on phishing, ransomware, and data breaches. About 300 instances were examined. The Hyperledger and Ethereum systems use SHA-256 and Keccak-256 hashing. The vetting method succeeded over 97% across all evidence categories. Specialist studies found a 98% agreement on the approaches' reliability. Blockchain technology creates verifiable, transparent, and immutable forensic records. Permissioned blockchains follow all rules and processes better than public networks. In addition, CipherTrace and Chainalysis Reactor analyzed 36 cryptocurrency samples. It was found that all Bitcoin, Ethereum, and stablecoin transactions could be tracked. Secret cryptocurrencies like Zcash and Monero have not been thoroughly investigated. The results support the idea that blockchain-based solutions improve digital evidence credibility and admissibility in court. This work's main contribution is a repeatable approach. The approach includes legal analysis, subject matter expert validation, and technological verification. Investigators and courts may verify digital artifact authenticity, traceability, and transparency using this technique.
Keywords:	Blockchain, Cyber Investigations, Digital Evidence, Forensic Authentication, Data Integrity, Cryptocurrencies, Legal Admissibility
DOI:	https://doi.org/10.5281/zenodo.19128500
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	THREATDETECTAI: AN AI-POWERED ZERO TRUST FRAMEWORK FOR REAL-TIME THREAT DETECTION AND CRYPTOGRAPHIC RE-VALIDATION IN CLOUD COMPUTING
Author:	D.VENKATESWARLU, Dr. B. SATEESH KUMAR
Abstract:	Cloud computing environments operate at a vast scale and support heterogeneous user activity, but ensuring integration is a key and significant security challenge. While a few attempts have been made to improve integrity verification, most existing integrity verification methodologies rely on a static cryptographic proof that involves an untrusted element or a rule-based validator. This independent anomaly detection method fails to detect minute insider threats, evolutionary attack patterns, or multi-modal correlations (e.g., log, network flow, and cryptographic traces). Although Zero Trust architectures employ many of these principles, the continuous validation of user-level trust and the prevention of information modification in a multi-tenant infrastructure create blind spots and limitations. This calls for an adaptive, intelligence-driven framework that integrates threat detection elements with cryptographic proof-checking. In this work, we introduce ZeroTrustAI, the first proof framework for AI-integrity, which combines a cryptographic verification engine with a hybrid deep learning algorithm called ThreatDetectAI. ThreatDetectAI utilises a CNN–BiLSTM–Attention model to identify spatial–temporal patterns in user behaviour and produce dynamic threat scores. Suspicious events are checked by Proofcryptnet (hash, digital signature, Merkle root). Such a system is integrated, providing a continuous, zero-trust verification loop with risk-based, adaptive access control. To evaluate the proposed approach, we conduct extensive experiments with many different data types, including cloud logs, network telemetry, and cryptographic metadata and show that it achieves 97.8% accuracy, 97.2% precision, 98.1% recall, and 0.986 AUC, which significantly outperformed the classical (RF, SVM, XGBoost) and deep learning baselines (CNN-only, LSTM-only, Transformer). The new version, with built-in cryptographic validation, achieved an 18% decrease in false positives when deployed for high-volume activity scenarios. We propose a framework that satisfies the scalability, explainability, and audit-friendliness criteria for unobtrusive, preemptive confidence assurance at the context-user level, thereby facilitating continuous authentication, adaptive authorisation, and strong, defence-grade countermeasures against adverse events in agile cloud settings. These results showcase the practical use of ZeroTrustAI to bolster protection in modern enterprise cloud environments.
Keywords:	Zero Trust Security, Deep Learning Threat Detection, Cryptographic Integrity Proof, Cloud Security Framework, User Behaviour Analytics
DOI:	https://doi.org/10.5281/zenodo.19133011
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	ENHANCED STATIC DEEPFAKE DETECTION THROUGH FACE SEGMENTATION AND TRANSFORMER-BASED MODELING
Author:	SAURABH KUMAR JAIN,MOHD AKBAR
Abstract:	The introduction of advanced generative models such as Generative Adversarial Networks (GANs) and diffusion architectures has made it easier to create hyper-realistic fake or tampered facial images. This advancement is very challenging to ensure the reliability of visual content in digital media. Detection of fake images is difficult and a major problem for issues related to privacy, security, and trust on the online platforms. The work proposed in this article demonstrates an integrated framework for real and fake face image detection that combines the advanced image preprocessing method with the Vision Transformer (ViT) based detection model. The image processing stage is applied to remove the backgrounds and isolate key facial regions. This process reduces unwanted objects and segments the discriminative facial features. The Vision Transformer restructures the face images into sequences of non-overlapping patches and applies global self-attention to capture subtle structural and textural inconsistencies that distinguish real and fake face images. Experimental results are presented that compare the performance of background removal, which helps to increase detection accuracy compared to unprocessed images. The proposed approach demonstrates strong performance on a very diverse dataset and successfully highlights the potential of transformer-based architectures for providing a scalable, interpretable, and robust deepfake detection method.
Keywords:	Deepfake, Vision Transformer, Face Detection, Image Segmentation, Data Mining
DOI:	https://doi.org/10.5281/zenodo.19133063
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	DEVELOPMENT OF AN AI-BASED PLATFORM FOR RECOMMENDING PROGRAMMING INSTRUCTION PLANS
Author:	SUPAPAT THANTURANON , SUWUT TUMTHONG
Abstract:	This research aims to develop and evaluate the effectiveness of AI-PINS (AI-based Programming Instruction Navigation System), a platform for recommending programming lesson plans for primary school teachers in municipal schools. It is designed within a six-theoretical framework: System Theory (IPOF), Agent Theory, Experiential & Adaptive Learning, Explainable AI (XAI), Information Theory, and Cognitive Load Theory. This platform consists of an AI-Agent Core that continuously recognises, analyses, and makes decisions based on real teaching data. The system applies DKT, CBR, RAG, Reinforcement Learning, and intelligent recommendation techniques to generate lesson plans that align with OBEC indicators and learner contexts. The evaluation consists of content validity (IOC ≥ 0.80), evaluation by 15 experts, statistical analysis of the mean, standard deviation, IOC, and Cronbach’s Alpha ≥ 0.90, and the application of TAM 4.0 to measure acceptability. The evaluation results indicated that the system is most appropriate (x ̅ = 4.67, S.D. = 0.13) especially in the dimensions of Intelligence Quality and Decision Quality (x ̅ = 4.65 and 4.80, respectively). This is due to the efficiency of ML, NLP, RAG, and the Explainable Decision module, which includes XAI to explain to users the reasons for recommending learning management plans and selecting activities through the Dashboard. Regarding the necessity of XAI in the education system, the system also has the potential for continuous learning through MLOps Monitoring and is user-friendly (x ̅ = 4.53), supporting TAM 4.0 factors in municipal schools. In summary, AI-PINS is a robust, well-evaluated architecture that helps reduce teachers' workload, supports personalised learning, and has the potential to expand digital learning policies at the national level.
Keywords:	AI-Based Platform, Artificial Intelligence, Learning Management Plan, Primary School Teacher, Programming Fundamentals
DOI:	https://doi.org/10.5281/zenodo.19133102
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	CAUSAL FOUNDATION MODELS FOR ONCOLOGY: A TEMPORAL MULTI-MODAL FRAMEWORK FOR COUNTERFACTUAL PROGNOSIS AND TREATMENT RESPONSE IN GLIOBLASTOMA
Author:	RAMAKRISHNA KOLIKIPOGU, Dr. K.V.VIVEKANANDA, CHOPPA.ANANDA KUMAR REDDY, THANU KURIAN, AMIT VERMA, Dr.R.SENTHAMIL SELVAN
Abstract:	Modern clinical AI systems are dominated by correlative models that predict outcomes from observed patterns but typically fail to provide reliable answers to interventional “what-if” questions required for treatment selection and policy evaluation. Such limitations reduce robustness when clinical practice, treatments, or patient subgroups shift. The study develops a causal multi-modal foundation model to estimate prognosis and treatment response for glioblastoma multiforme (GBM), enabling counterfactual reasoning over longitudinal treatment timelines. The present study compares three model classes: (i) Baseline - a standard multi-modal transformer using pre-trained encoders (UMME); (ii) Advanced - a causal multi-modal foundation model that injects structural causal constraints and causal regularization during training; and (iii) Proposed - a Temporal Causal Multi-Modal Transformer (TCMMT) that models temporal treatment patterns, integrates dynamic multi-modal diagnostics, and contains a counterfactual head for intervention simulation. Training data were drawn from TCGA-GBM and longitudinal institutional cohorts with curated treatment timelines and censoring-aware labels. The proposed TCMMT achieved a concordance index (C-index) of 0.87 for survival prediction and improved counterfactual treatment response prediction accuracy by 23% relative to non-causal baselines (relative improvement). By combining causal graph structure, temporal encoding of treatment histories, and multi-modal foundation encoders, TCMMT supports in-silico clinical trials and personalized treatment optimization, marking a shift from purely predictive to prescriptive AI in oncology.
Keywords:	Causal AI, Foundation models, Temporal multi-modal learning, Counterfactual reasoning, Glioblastoma multiforme (GBM), Structural causal model (SCM).
DOI:	https://doi.org/10.5281/zenodo.19133154
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	IOMT- BASED SYSTEM FOR EARLY DETECTION OF CARDIAC ISSUES: CORE FEATURES AND DESIGN
Author:	VIDHYA G , VIJAY BHANU
Abstract:	Cardiovascular diseases are one of the main reasons for the deaths in the world, many people mostly because of the early or short-time changes in heartbeat are not going to identified quickly. The Internet of Things (IoMT) helps us in watching the patient’s heart condition continuously, but the thing is in the existing systems face a lot of problems like delay in giving real-time results, weak data protection, and also high-power use in wearable devices. In this work, we are going to introduce a secure edge-based IoMT system that can find heart problems early by using ECG data. The model uses a small and efficient CNN so that can work even on low power devices and can detect arrhythmia, tachycardia, and bradycardia within one second since the processing happens at the edge instead of the cloud. To keep the data safe, we have used AES and TLS encryption, which follow HIPAA and GDPR privacy rules. The main goal is to make the system practical by keep a proper balance between the speed, security, and energy saving. Unlike the systems that depend too much on cloud or blockchain, this method works near the patient, so it gives fast results with less delay of time. By comparing the quick edge processing, lightweight machine learning, and strong encryption, the above work builds a base for future IOMT systems that are safe, reliable, and ready for real-world use.
Keywords:	Cardiovascular Diseases, Internet of Things (IoMT), Wearable Devices, Encryption, Blockchain
DOI:	https://doi.org/10.5281/zenodo.19133190
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	DEEP LEARNING-DRIVEN FLEXIBLE BIOSENSOR SYSTEM FOR CONTINUOUS HEALTH MONITORING AND EARLY DISEASE DETECTION
Author:	SIVA SANKAR NAMANI, Dr. TAVITI NAIDU GONGADA, CHOPPA.ANANDA KUMAR REDDY, Dr. MAHAVIR A. DEVMANE, AMIT VERMA, Dr.R.SENTHAMIL SELVAN
Abstract:	There is a fundamental accuracy efficiency trade-off between continuous low-burden physiological monitoring with flexible patches: server-bound clinical-grade models, and on-device methods of achieving sensitivity to meet energy constraints. To obtain diagnostic performance on microcontroller-class wearables similar to that of servers, this study presents a Conv-Transformer-GNN architecture that is ready for deployment. The system incorporates multimodal sensor fusion, knowledge distillation, and federated aggregation. The main achievement is the proof that graph-based multimodal fusion with quantization-aware distillation may maintain clinical accuracy while fitting within tight memory, latency, and energy budgets. The distilled edge model closely follows the instructor in terms of experimental outcomes (ROC-AUC 0.89 vs. 0.92), while having a small footprint of 340 KB, an inference energy of approximately 4.5 mJ, and a p95 latency of about 85 ms. These results provide useful guidelines for the development of adaptable biosensor systems that can scale while protecting users' privacy.
Keywords:	Flexible Patch; Edge Computing; Conv-Transformer-GNN; Federated Aggregation; Knowledge Distillation.
DOI:	https://doi.org/10.5281/zenodo.19133210
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	BUILDING USER-CENTRIC AGENTIC SYSTEMS: DESIGN PRINCIPLES AND INTERFACE CONSIDERATIONS FOR AI-DRIVEN DECISION SUPPORT
Author:	RAJA MOHD TARIQI RAJA LOPE AHMAD, MUHAMMAD FAIRUZ ABD RAUF, RITA WONG MEE MEE, NUR AMLYA ABD MAJID, SHARIFAH AISHAH SYED ALI, FAZILATULAILI ALI, ZURAINI ZAINOL2, MOHD FAHMI MOHAMAD AMRAN
Abstract:	Artificial intelligence systems are capable of analysing complex data and generating valuable insights, yet their impact is reduced when users are unable to understand or trust the recommendations produced. This paper introduces a user centred approach to the design of artificial intelligence interfaces that aim to make complex insights both accessible and trustworthy. Using the AutoAgentic financial analysis platform as a case study, we illustrate how careful interface design can connect advanced artificial intelligence capabilities with practical human usability. Through an iterative process of design and evaluation, interface solutions were developed and tested to support the visualisation of market regimes, explanation of system recommendations, communication of confidence levels and provision of appropriate user control. The results indicate that visual simplicity, layered information presentation, plain language explanations and meaningful user control substantially enhance user understanding and trust. Based on these findings, four reusable design patterns are proposed, namely Confidence Thermometer, Why This Recommendation, Progressive Disclosure and Plain Language Translator, which can be applied across a wide range of artificial intelligence decision support systems. This study contributes practical guidelines for the design of artificial intelligence interfaces that effectively support human decision making through transparent, understandable and trustworthy interactions.
Keywords:	User Interface Design, Explainable AI, Decision Support Systems, Human-AI Interaction, Financial Analytics, Agentic Systems,
DOI:	https://doi.org/10.5281/zenodo.19133230
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	USE OF DIGITAL PUBLIC GOVERNANCE TOOLS IN A STATE OF EMERGENCY TAKING INTO ACCOUNT ADMINISTRATIVE AND LEGAL RESTRICTIONS
Author:	IRYNA MURAVIOVA, OLEKSANDR YERMAK, VIKTORIIA KORETSKA, VALERIIA RIADINSKA, NADIIA VASYLENKO
Abstract:	The article presents the results of a comparative study of the use of digital public governance tools in a state of emergency using the example of Ukraine, Germany, and Estonia. The aim is to assess the administrative and legal balance of digital governance using the integrated Administrative and Legal Balance Index (ALBI), which covers efficiency (Eff), transparency (Transp), legal compliance (LegComp), and human rights protection (HRProt). The methodology combines comparative law, statistical and modelling approaches using k-means clustering, PCA visualization, bootstrap estimation (n = 1000), and scenario modelling of three management situations (interdepartmental coordination, electronic identification, citizens’ appeals). The ALBI weights were determined based on the results of a three-round Delphi survey and AHP-consensus of 27 experts from three countries. The research is based on the analysis of regulatory acts (Onlinezugangsgesetz, Public Information Act, Government Regulation No. 105; the laws of Ukraine “On the Legal Regime of Martial Law”, “On Electronic Trust Services”) and the aggregated indicators DESI, EGDI, and Rule of Law Index. It was found that Germany has the highest level of balance (ALBI = 0.86) due to the combination of technological maturity with effective judicial control. Estonia demonstrates maximum efficiency (Eff = 0.88) and regulatory stability due to X-Road and the data once only principle. Ukraine is characterized by high rates of digitalization (Eff = 0.78) with lower values of transparency (Transp = 0.61) and human rights protection (HRProt = 0.59). It was established that the balance of digital governance critically depends on harmonization with Regulation (EU) 2024/1183 (eIDAS 2.0), Recommendation CM/Rec(2018)7 of the Council of Europe, and the OECD Digital Government Framework. The academic novelty is the developed lawful-by-design model that integrates digital solutions with administrative and legal procedures at the design stage. Further research prospects for are related to the creation of adaptive crisis management systems capable of automatically maintaining a balance between speed, legality, and protection of citizens’ rights in emergency legal regimes.
Keywords:	Digital Public Governance, E-Governance, Administrative Law, Rule Of Law, Human Rights, Sustainable Institutions
DOI:	https://doi.org/10.5281/zenodo.19136512
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	7G WIRELESS NETWORKS: HARNESSING TERAHERTZ-TO-LIGHTWAVE FOR ULTIMATE CAPACITY
Author:	SUBRAMANYA SARMA S, DR D. NAGARAJU, BHUVANESWARI S, PRAVEEN KUMAR,GOPAGON, K.DEEPTHI, VEERASWAMY AMMISETTY, DR.SURESH KUMAR, PITTALA , DR.T.VENGATESH, M.L.M.PRASAD, DR.BH.KRISHNA MOHAN
Abstract:	The exponential growth in data traffic, driven by immersive applications like holographic communications, volumetric video, and the tactile internet, is rapidly pushing the Shannon limit of current sub-6 GHz and millimeter-wave (mmWave) spectrums. This paper explores the foundational role of Terahertz (THz: 0.1-10 THz) and Light wave Communication (Li-Fi, Free-Space Optical - FSO) bands as the cornerstone for the seventh-generation (7G) wireless networks. We propose a novel, hierarchical network architecture that seamlessly integrates THz bands for ultra-dense, short-range access and light wave carriers for high-capacity backbone and front haul links. A comprehensive literature survey establishes the current state-of-the-art, highlighting the complementary strengths and limitations of both technologies. We then detail a proposed system methodology, including a hybrid beam forming and intelligent reflecting surface (IRS) assisted THz system, coupled with a multi-wavelength FSO backbone. Simulation results, based on realistic channel models and datasets, demonstrate that the proposed THz-to-Light wave architecture can achieve aggregate data rates exceeding 1 Tbps per access point and latency below 100 μs, thereby supporting the key performance indicators (KPIs) envisioned for 7G. The paper concludes with a discussion on implementation challenges, including mobility management and atmospheric effects, and outlines a path forward for realizing the ultimate capacity of wireless networks
Keywords:	7G, Terahertz (THz), Light wave Communication, Li-Fi, FSO, Hybrid Beam forming, Intelligent Reflecting Surface (IRS), Tbps
DOI:	https://doi.org/10.5281/zenodo.19136530
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	DIGITAL AND AI TOOLS IN STATE FINANCIAL POLICY: EUROPEAN EXPERIENCE AND A CASE FOR UKRAINE
Author:	HENNADII ZHUYKOV, SVITLANA PROKHORCHUK, DMYTRO MAZUR, YURI KULYNYCH, HALINA MAZUR, OLEKSANDR MAZUR
Abstract:	In the current conditions of global digital transformation, public finances face challenges and opportunities associated with the introduction of digital technologies and artificial intelligence (AI). The relevance of the study is due to the need to assess how digitalization and AI affect the effectiveness of financial policy in different countries, especially in the context of the challenges that Ukraine faces when reforming public finances in a period of economic and social change. After all, a high level of digital services and the active use of e-government do not always translate into increased fiscal transparency, budget discipline, and the effectiveness of financial policy. The article carries out a comprehensive interdisciplinary analysis of the impact of digital instruments on financial policy using the example of Ukraine, its neighbors from the European Union (Poland, Slovakia, Hungary, Romania) and Turkey, a candidate country with active digitalization. The methodology is based on a systematic combination of quantitative and qualitative indicators - from the share of the ICT sector in gross value added, the volume of public and private funding of scientific research, the level of digital skills of the population and the use of AI, to indices of e-government development and transparency of the budget process. Particular attention is paid to imbalances between the digital skills of the population, the institutional capacity of the state, the innovative activity of business and the quality of budget management. The results revealed significant differences in the level of digital readiness, innovation potential and institutional capacity of countries to use digital technologies in financial policy. Ukraine is characterized by high demand for digital services, but has limitations in the human resources of researchers and weak institutional support for transparency and citizen participation. The comparative analysis confirmed that to increase the effectiveness of financial policy, a transition from the simple provision of digital services to a comprehensive institutional transformation that integrates AI and digital tools into all stages of the budget cycle is necessary. In addition, the most effective impact of digitalization on financial policy is observed in countries with a high level of digital skills, a developed R&D infrastructure, a powerful ICT sector, and balanced indicators of budget transparency and citizen participation (Poland, Slovakia). On the other hand, in Ukraine, despite the high digital activity of citizens and the dynamic development of e-government, the effect of digitalization is restrained by insufficient institutional capacity, a low level of budget transparency, a weak innovation sector and a decrease in human scientific and technological potential. The practical significance of the work lies in the formulation of recommendations for Ukraine on the adaptation of European models of digital transformation of finance, which can become the basis for building a transparent, accountable and efficient financial system. The study is also useful for other countries with similar socio-economic challenges that seek to integrate innovations into public financial management.
Keywords:	Digitalization, Artificial Intelligence, Financial Policy, E-Government, Digital Skills, Budget Transparency, Institutional Capacity, Public Finances, Ukraine, European Union, Turkey.
DOI:	https://doi.org/10.5281/zenodo.19136567
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	INTEGRATION OF TECHNOLOGIES INTO THE PRIMARY SCHOOL LEARNING PROCESS: BENEFITS AND CHALLENGES
Author:	VIKTORIIA PAVELKO, ALINA PREDYK, IRYNA RADCHENIA, YULIA RYABOKIN
Abstract:	The need to integrate modern innovative tools into the formation of primary school students’ competencies dictates the need to change approaches and tools for the educational process. The goal is to analyze the current challenges and advantages of using digital technologies in primary schools based on theoretical research and practical tools. Research methods. The paper considers innovative pedagogical tools that have been developed based on modern digital solutions, in particular online platforms and other interactive tools, as well as blended learning technologies for primary school age children. An experiment was conducted using pedagogical experimentation and virtual technologies. The results of the experiment were used in writing this study. In particular, emphasis was placed on the main advantages of innovative technologies that contribute to increasing the interest of young students, facilitate the learning process, and improve the quality of information assimilation. At the same time, interactive technologies allow for the most effective study of educational material in inclusive learning spaces. However, it is worth noting the possibility of information overload, decreased attention and concentration, increased risks of negative effects on children’s health, and deterioration of social interaction. It has been proven that the use of visualization tools makes it possible to explain complex circumstances and abstract concepts to children, which improves the effectiveness of perception, engages the sensory and emotional spheres, and promotes critical thinking.
Keywords:	Interactive Technologies, Innovations, Natural Sciences, Future Teachers, Primary Education, Teaching Methodologies
DOI:	https://doi.org/10.5281/zenodo.19136592
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	REAL-TIME CONTEXT AWARE VARIABLE STRENGTH CRYPTOGRAPHY FOR CLOUD STORAGE
Author:	Dr.A.BARAKATH BEGAM, Dr.A.BHUVANESHWARI, Dr.J.JENIFER, Dr.S.VEERAPANDI, Dr.S.MURUGANANDAM, PRAJWALASIMHA S N, POTHUMARTHI SRIDEVI, Dr.L.THENMOZHI, DR.T.VENGATESH
Abstract:	In response to the escalating dependence of individuals and enterprises on cloud storage solutions, the demand for robust data protection mechanisms has become increasingly imperative. Conventional cryptographic techniques typically use a fixed level of encryption strength, which can either waste resources or fail to provide adequate protection by not adjusting to different levels of data sensitivity or evolving security threats. This paper introduces an innovative solution called Real-time Context-Aware Variable Strength Cryptography (RTCAVSC) for cloud storage. RTCAVSC automatically modifies encryption strength in real time based on various contextual factors, such as how sensitive the data is, patterns in user access and the prevailing security conditions. Fuzzy Real-time Sensitivity Tracer, Dynamic Cloud Environment Interpreter and Variable Strength Cryptography Manager are the novel contributed modules assembled unison in RTCAVSC work. A dedicated cloud server is leased to evaluate the performance of RTCAVSC method in terms of Encryption Time, Decryption Time, Throughput, Latency, Resource Utilization and Data Security parameters in real-time environment.
Keywords:	Context Aware, Cloud Data Security, Real-Time Variable Strength Cryptography, Variable Strength Cryptography
DOI:	https://doi.org/10.5281/zenodo.19136602
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	XMGNET: A CROSS-MODALITY GRAPH NEURAL NETWORK WITH MODALITY-AWARE RESIDUAL ALIGNMENT FOR MULTIMODAL EMOTION RECOGNITION
Author:	RAMAPRAKASH KALAPALA, MAHESH YADLAPATI, S. VIJAYA NIRMALA, VINOD GOJE, VIJAYA BHASKAR SADHU, DR. B. NARENDRA KUMAR
Abstract:	Emotion estimation from multimodal information is now the focus of affective computing, especially for mental health applications, adaptive user interfaces, and human-robot interaction. Fusion of multimodal biosignals, though, is still a main concern owing to varying temporal patterns, varying levels of signal quality, and the lack of practical inter-modal alignment techniques. This work proposes XMGNet (Cross-Modality Graph Neural Network), a new graph-attentional model that represents emotion as an intermodal joint function of spatial, temporal, and semantic relationships between modalities. Our model builds dynamic graphs for each of the modalities—EEG, ECG, GSR, and facial expressions—prior to modality-wise graph convolutions blended with multi-head cross-attention for hierarchical fusion. To enhance robustness and generalization, we present a Modality-Aware Residual Alignment (MARA) block that adjusts at runtime to missing or corrupted channels. We test XMGNet on three publicly released datasets: DEAP, DREAMER, and MAHNOB-HCI, with state-of-the-art accuracy of 94.3%, 91.7%, and 89.8%, respectively, performing better than recent transformer- and LSTM-based models. The proposed model exhibits scalable, explainable, and high-performance emotion recognition without the need for handcrafted synchronization, with potential deployment on real-time emotion-aware systems.
Keywords:	Multimodal Emotion Recognition; Graph Neural Networks; Cross-Modality Fusion; EEG; ECG; GSR; Facial Expressions; Modality-Aware Alignment; Affective Computing; Attention Mechanisms
DOI:	https://doi.org/10.5281/zenodo.19136622
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	TRI-PARADIGM ANALYSIS OF COMPUTATIONAL COMPLEXITY IN NP-HARD SUDOKU SOLVERS USING EXPLAINABLE AI
Author:	RAJAN THANGAMANI, PALLAVI R
Abstract:	Sudoku is a demanding test of computational complexity and constraint satisfaction problems. This work develops a comprehensive and replicable benchmark that reformulates 9×9 Sudoku puzzles under three solver paradigms of constraint programming (CP), Boolean satisfiability (SAT), and integer programming (IP) in order to assess efficacy and explainability. For each paradigm, a parametrically equivalent model is implemented and solved with common datasets and experiment control conditions. We measure wall-clock running time, peak memory usage, and paradigm-data related measures of effort in backtracks or fails in CP, decisions or conflicts with conflict-driven clause learning(CDCL) in SAT, and branch & bound nodes or linear program relaxations in IP. Aggregate trends provide empirical evidence consistent with NP-hard/NP-complete behavior: runtime grows approximately exponentially with constraint tightness, while memory remains nearly flat for standard grids. An explicit Explainable Artificial Intelligence (XAI) trade-off emerges. SAT delivers the lowest runtime via clause learning and pruning; CP yields the most interpretable reasoning through transparent propagation and traceable search; IP offers algebraic auditability and flexibility for optimization variants but incurs higher search effort on pure feasibility. The benchmark, figures, and tables establish a verifiable claim–evidence link and supply a practical baseline for integrating explainability metrics into analytic decision pipelines. The results inform the design of intelligent and explainable solver architectures, aligning with the aim of Intelligent systems with applications. We conclude with a roadmap toward hybrid CP–SAT approaches that retain CP-level interpretability while leveraging SAT-level efficiency.
Keywords:	Sudoku, NP-Complete, Constraint Programming, SAT, Integer Programming, Explainable AI
DOI:	https://doi.org/10.5281/zenodo.19136632
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	INFORMATION TECHNOLOGIES AS A MEANS OF IMPLEMENTING PUBLIC POLICY IN THE FIELD OF E-DEMOCRACY
Author:	LILIIA BURACHOK, IVANNA LOMAKA, OKSANA LYPCHUK, RUSLAN IBRAHIMOV, NATALIIA SHOTURMA
Abstract:	Information technologies are revolutionizing the landscape of contemporary public policy by facilitating extensive communication avenues between the state and its citizens. The purpose of this study is to systematically evaluate the impact of e-government development on the quality, effectiveness, and accountability of public administration, as well as its role in reducing corruption, using empirical methods to identify which aspects of digital governance exert the strongest influence. The research employed methodologies such as multiple linear regression, ANOVA, correlation analysis, as well as radar chart analysis. This paper investigates the impact of e-government on the quality of public administration, specifically focusing on the indicators "Voice and Accountability", "Government Effectiveness", and "Control of Corruption". An analysis encompassing 193 countries worldwide was conducted utilizing data from 2014 to 2022, drawing upon three sub-indices of the E-Government Development Index: the Human Capital Index, the Online Services Index, and the Telecommunications Infrastructure Index. The findings revealed that the Voice and Accountability indicator is significantly positively influenced by both the Human Capital Index and the Telecommunications Infrastructure Index, with Beta regression coefficients of 0.35 and 0.25, respectively. Conversely, the Online Services Index did not emerge as a significant determinant. These results underscore the critical importance of human capital development and telecommunications infrastructure in enhancing the efficacy of public administration and curtailing corruption. In contrast, online services exhibit a relatively weak impact unless synergized with other factors. The practical implications of this research suggest that its findings could contribute to improving strategies for the digital transformation of public administration through the advancement of human capital and telecommunications infrastructure. Future research should concentrate on comparing nations with varying levels of digital advancement to discern localized governance characteristics and examine the adverse effects of digitalization on democracy.
Keywords:	Information Technology, Public Policy, E-Government, E-Democracy, Telecommunications Infrastructure
DOI:	https://doi.org/10.5281/zenodo.19136654
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	A DESIGN FRAMEWORK FOR PERSONALIZED INTELLIGENT TUTORING SYSTEMS INTEGRATING KNOWLEDGE TRACING AND RETRIEVAL-AUGMENTED GENERATION
Author:	POONNAKAN CHAMNANKIJ, EKACHAI NAOWANICH, SUWUT TUMTHONG
Abstract:	This research presents a study on the design framework of an intelligent, personalized tutoring system to meet the individual needs of learners, overcoming the limitations of traditional learning management systems (LMS) that lack flexibility and brainstorming. The researchers designed a system architecture based on the Intelligent Tutoring System (ITS) concept, comprising seven main modules such as pre-assessment, intelligent learner grouping (AI-Classify), and blended learning, built on a cloud-based database structure. The developed system applies advanced techniques, including Retrieval-Augmented Generation (RAG) to increase accuracy and reduce AI hallucinations; Knowledge Tracing (KT) for continuous analysis of learner knowledge status; and an AI Chatbot to act as a continuous academic advisor. Expert evaluation of the conceptual framework's consistency showed an Index of Content Validity (IOC) for each module ranging from 0.89 to 0.97, with an average of 0.93, exceeding the acceptable criterion (≥ 0.50), indicating the appropriateness and consistency of the content with the research objectives. The overall system suitability assessment was at the highest level (x̅ = 4.71, S.D. = 0.45), and the technical suitability dimension of the AI innovation had a high average score of 4.73. In conclusion, this learning support system is designed to support deep learning, analytical thinking, and learner participation, with the potential to enhance educational practices aligned with 21st-century skills.
Keywords:	Intelligent Tutoring Systems, Personalized Learning, Knowledge Tracing, Retrieval-Augmented Generation, Artificial Intelligence in Education
DOI:	https://doi.org/10.5281/zenodo.19136667
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	TRANSFORMER–GNN–UNET HYBRID DEEP LEARNING FOR SCALABLE WIRELESS RADIO MAPS
Author:	DR. SASIPRIYA G, N VISWANADHAREDDY, CHOPPA.ANANDA KUMAR REDDY, Dr.M.SUNDAR RAJAN, AMIT VERMA, Dr.R.SENTHAMIL SELVAN
Abstract:	Wireless environments are complex and require high-resolution radio maps for tasks such as indoor localization, spectrum management, and automated planning. Under severe sparsity, empirical models have trouble preserving fine structural information and scale inadequately. This paper proposes a Transformer–GNN–UNet hybrid deep learning approach for scalable wireless radio maps: it fuses global attention (Transformer), topology-aware reasoning (GNN), and multiscale convolutional reconstruction (UNet) with physics-informed priors and probabilistic outputs. Models were evaluated on the Indoor Radio Map Benchmark and the MLSP 2025 dataset across sampling regimes (0.02%, 0.5%), using RMSE, SSIM, LPIPS, CRPS, and topological (bottleneck) metrics. The hybrid achieves RMSE 3.45 dB (95% CI ±0.12) versus 4.18 dB for SAIPP-Net and 6.84 dB for the 3GPP baseline; SSIM improves to 0.912 (vs 0.876), and CRPS drops to 0.95 (vs 1.26); bottleneck distance falls to 0.035 (vs 0.048). Results are statistically significant (p = 0.001) and consistent across seeds; runtime (~48 ms) supports near-real-time use. The proposed Transformer–GNN–UNet hybrid offers a scalable, calibrated, and deployable solution for high-fidelity wireless radio mapping.
Keywords:	Transformer–GNN–UNet, Wireless Radio Maps, Hybrid Deep Learning, Scalable Pathloss Prediction, Uncertainty Calibration.
DOI:	https://doi.org/10.5281/zenodo.19136688
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	FORECASTING SAUDI TOURISM DEMAND FOR VISION 2030: A COMPARATIVE ANALYSIS OF REGRESSION AND LSTM MODELS
Author:	MASHEL AL-FEHAID , NAHALH ALGETHAMI
Abstract:	Tourism is central to Saudi Arabia’s Vision 2030 diversification agenda. Reliable demand intelligence is needed to plan investment, capacity, and seasonally targeted interventions. Through substantial investments in infrastructure, heritage conservation, and digital transformation, Vision 2030 aims to position Saudi Arabia as a leading global tourism destination. Existing studies rarely integrate driver attribution with robust nonlinear forecasting, and often under-capture seasonal and pandemic shocks. To address this gap, we develop a dual-scope framework that (i) explains variation in total tourists using regression and (ii) forecasts future inflows with deep learning. The driver model examines tourist type, reason for visit, spending behavior, religious periods (e.g., Ramadan, Hajj), and pandemic effects. A degree‑3 polynomial regression achieves the highest explanatory fit (R² = 0.805), with reason for visit and tourist type emerging as significant determinants. For forecasting, we design and compare several LSTM architectures; the best, a generator–discriminator LSTM, attains testing R² = 0.756, capturing nonlinear and seasonal structure. Together, these results unite driver evidence with accurate forecasts, supporting targeted marketing, resource allocation, and season-specific operations in line with Vision 2030. The study contributes an integrated dual scope framework for Saudi arrivals, domain specific exogenous design such as religious/pandemic, and comparative deep learning benchmarks aligned to policy relevant metrics moving beyond determinants only studies, spending centric ML/ARIMA, and attribution free ensembles. By uniting interpretable driver evidence with accurate deep‑learning forecasts under Vision 2030, this study offers decision‑grade demand intelligence that the current literature has not yet provided in a single, Saudi‑specific framework.
Keywords:	Machine Learning, Deep Learning, Regression Analysis, Long Short-Term Memory, Tourism Sector.
DOI:	https://doi.org/10.5281/zenodo.19136702
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	DEVELOPMENT AND EVALUATION OF A CONCEPTUAL FRAMEWORK FOR AN INTELIGENT PROJECT TOPIC RECOMMENDATION AND TIOMELINE SYSTEM IN UNDERGRADUATE PROJECT COUTSES TO ENHANCE ALIGNMENT WITH COURSE ENHANCE ALIGNMENT WITH COURSE LEARNING OUTCOMES
Author:	ANGSANA PHONSUK , SUWUT TUMTHONG
Abstract:	Undergraduate capstone project development faces critical challenges, including students' difficulty in selecting topics aligned with course learning outcomes, inefficient time management, and prolonged topic approval processes that delay graduation timelines. This research develops and evaluates a conceptual framework for an intelligent project topic recommendation and timeline management system integrating Constructive Alignment theory, artificial intelligence, and project management principles. The framework comprised five core components data management and integration,intelligent recommendation system (Hybrid techniques), natural language processing for topic-CLO analysis, automated timeline planning and continuous evaluation Expert evaluation by nine specialists across five dimensions showed highly favorable results (mean = 4.50, SD = 0.56), with System Quality (M = 4.56), Intelligence Quality (M = 4.44), Decision Quality (M = 4.52), Learning Quality (M = 4.44), and User & Environment Quality (M = 4.56) all achieving "High" to "Highest" appropriateness levels. The developed framework has significant potential to address capstone project challenges and is feasible for practical implementation in educational institutions.
Keywords:	Project Topic, Recommendation, Timeline System, Chatbot, Tracking, Course Learning Outcomes
DOI:	https://doi.org/10.5281/zenodo.19136729
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	INTEGRATION OF ARTIFICIAL INTELLIGENCE IN PUBLIC GOVERNANCE OF INTELLIGENT TRANSPORT SYSTEMS
Author:	GRISHA AMIRKHANYAN, NATALIIA PROTSIUK, VLADYSLAV KARIEV, OLEKSANDR FOMIN, PAVLO SOLOVII
Abstract:	Relevance of the research The relevance of the study is determined by the need to implement cognitive and adaptive artificial intelligence (AI)-based solutions to increase the efficiency, sustainability, and regulatory manageability of intelligent transport systems (ITS) in public governance. Research objective The research objective is to substantiate the methodological principles of creating an AI-architecture of ITS in public administration, taking into account cognitive adaptability, interoperability, and regulatory consistency. Research methods The research employed the following methods: retrospective stratification, metric-based and model-based analysis, Unified Modelling Language (UML)-based modelling, structural and functional optimization, metric-based and model-based verification. Results A multi-level verification of the AI-Optimized ITS Framework for Public Governance was carried out through retrospective stratification, multimetric analysis, UML-based modelling and empirical indicator assessment. Cognitive inertia, institutional fragmentation and regulatory decentralization of traditional ITS were identified. Metrically confirmed efficiency increase with TTR=0.21, QL=0.25, Accuracy=0.93, F1=0.92, ΔCO₂=0.90, FC=0.92, RI=0.89, CRI=0.91, ATS=0.94. AI architecture demonstrates regulatory traceability, cognitive adaptability and management scalability. Academic novelty of the research The academic novelty of the research is the formalization of the cognitive and adaptive AI-Optimized ITS Framework for Public Governance with the architecture of Explainable AI (XAI)-transparency, regulatory traceability, and algorithmic optimization. The integration of cognitive interoperability, regulatory validation, and indicator efficiency in smart mobility management is ensured for the first time. Further research prospects Future research prospects include initiating a controlled field experiment in urban transportation systems to collect validated field data. This approach will provide an empirical assessment of the algorithmic accuracy, procedural resilience, and normative traceability of the proposed framework.
Keywords:	AI-Optimized ITS, Cognitive Adaptivity, Explainable AI (XAI), Travel Time Reduction, Queue Length Optimization, CO₂ Emission Minimization, Smart Urban Mobility
DOI:	https://doi.org/10.5281/zenodo.19136753
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	FEDERATED DIFFERENTIALLY PRIVATE AUTOENCODERS FOR HEALTH-INSURANCE FRAUD DETECTION AT SCALE
Author:	GANESH SHANKAR SARGAM , RAMPRAKASH KALAPALA
Abstract:	Health-insurance fraud has grown in scale and sophistication, imposing substantial financial losses and undermining trust in healthcare systems. Conventional, centrally trained detectors require aggregating sensitive claim data, increasing breach risk and complicating regulatory compliance. Despite advances in fraud analytics, a core unresolved challenge remains: enabling multi-institution collaboration without exposing raw claims data while preserving detection accuracy. We present a federated anomaly-detection framework that combines client-side deep autoencoders with Federated Averaging (FedAvg) and differential privacy (DP), orchestrated on Amazon Web Services (AWS) for secure, scalable deployment. Participating organizations train locally on private claims and transmit only DP-protected model updates to a cloud aggregator (EC2/SageMaker), while encrypted artifacts are persisted in Amazon S3. Using real insurance-claim records augmented with realistic synthetic fraud cases, the proposed system achieves 91.2% accuracy and an F1-score of 0.885, and reduces false-positive rate by 43% relative to a traditional centralized baseline. Privacy controls lower estimated data-exposure risk by 78% with minimal communication overhead, enabling near real-time operation across multiple sites. These results indicate that federated training with formal privacy protection can maintain strong detection performance while limiting sensitive data movement. This study contributes empirical evidence that integrating differential privacy within federated unsupervised anomaly detection can sustain high detection performance under heterogeneous data distributions while providing quantifiable privacy–utility trade-offs. Overall, the framework offers an effective, scalable, and privacy-aware solution for distributed fraud detection, supporting compliance with regulations such as HIPAA and GDPR and facilitating collaborative analytics without raw-data sharing.
Keywords:	Federated Learning; Differential Privacy; Autoencoders; Anomaly Detection; Health-Insurance Claims; Fraud Detection; Secure Aggregation; Cloud Orchestration (Aws); Decentralized Machine Learning.
DOI:	https://doi.org/10.5281/zenodo.19136763
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text

Title:	A HYBRID DEEP LEARNING-BASED PATH-PLANNING ALGORITHM FOR MULTIFUNCTIONAL MANIPULATOR-TYPE DISINFECTION ROBOT
Author:	DR. KOVVURI N BHARGAVI, TANAYA GANGULY, MANOJ KUMAR PADHI, DR.G. JOSE MOSES, DR. NIRAJ KUMAR, NAGENDAR YAMSANI
Abstract:	This research focuses on developing an advanced path-planning algorithm for a multifunctional manipulator-type disinfection robot, incorporating deep learning techniques to optimize its operational efficiency and adaptability. Moreover, these robots are used to disinfect surfaces in sensitive environments, such as public facilities, laboratories, and hospitals. However, optimizing the performance of robots is challenges because energy usage, and disinfection coverage time can make dynamic environments and navigating difficult. Also, arm co-ordination and obstacle avoidance do not provide effective disinfection due to the changing criteria. To overcome these issues, the developed study reveals that the hybrid solution which is the integration of Deep learning (DL) frameworks and path planning strategies enhances efficiency as well as adaptability. The study's objective is to address the limitations of traditional algorithms, like A* and Dijkstra’s, which fail in dynamic environments such as hospitals where obstacles are constantly changing. Moreover, the developed algorithm finds the better route from the starting location of robot to the robot target while considering the difficulty and map structures. To overcome these limitations, a hybrid model is proposed, combining the A* algorithm for global path planning with Proximal Policy Optimization (PPO), a deep reinforcement learning method for real-time path adjustments. The methodology involves training the robot in simulated environments and testing it in real-world settings such as hospitals. Results show significant improvements in path efficiency, energy consumption, and collision avoidance, with the proposed algorithm achieving 98.5% coverage efficiency. The implications of this research highlight the potential of deep learning to enhance the adaptability and effectiveness of disinfection robots, especially in dynamic, high-traffic environments, ensuring thorough sanitation and minimizing energy usage.
Keywords:	Path Planning, Deep Reinforcement Learning, Disinfection Robots, A* Algorithm, Proximal Policy Optimization.
DOI:	https://doi.org/10.5281/zenodo.19136782
Source:	Journal of Theoretical and Applied Information Technology 15^th March 2026 -- Vol. 104. No. 5-- 2026
Full Text