		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
October 2026 | Vol. 104 No.19

Title:	A MULTI-MODEL SOFT-VOTING HYBRID ENSEMBLE DEEP LEARNING FRAMEWORK FOR ADVANCED TEXT CLASSIFICATION
Author:	D.JASMINE GUNASUNDARI , J.ARUNADEVI
Abstract:	This study addresses a specific difficulty in automated emotion recognition from social media text: informal spelling, slang, elongated words, emojis, and hashtags routinely degrade classifier accuracy. A Multi-Model Deep Learning Architecture for Social Media Emotion Detection (MMDLA-SMED) is proposed, combining a dedicated normalization pipeline with a soft-voting ensemble classifier. Raw posts are processed through cleaning, hashtag segmentation, slang expansion, emoji-to-text conversion, word-elongation normalization, and lightweight spelling correction. The normalized text is encoded using FastText, whose sub-word character n-gram representation embeds out-of-vocabulary and misspelled terms common in social media posts. Classification combines three heterogeneous deep learning architectures, an improved temporal convolutional network (TCN), a residual gated recurrent unit (ResGRU), and a hierarchical attention network (HAN), trained independently and merged through soft voting so convolutional, recurrent, and attention-based decision patterns jointly determine the predicted emotion. MMDLA-SMED is evaluated on two benchmark tweet-emotion datasets, one with 12 classes and 39,173 samples and another with 6 classes and 16,000 samples, under a 70:30 train-test split. On the 12-class dataset, the model attains a testing accuracy of 98.72%, precision of 87.95%, recall of 73.15%, and F1-score of 74.82%; on the 6-class dataset it attains a training-phase accuracy of 98.38%, precision of 94.28%, recall of 91.56%, and F1-score of 92.81%, outperforming Random Forest, Decision Tree, LightGBM, GCN, GraphSAGE+Pooling, GoLA, DIMINANT, GAAN, 1D-CNN, LSTM, Bi-LSTM, CNN-LSTM, CNN-BiLSTM, MAPLE, and DSGN on every measure. Class-wise results show minority categories such as boredom and anger are recognized with markedly lower recall than dominant categories, reflecting class imbalance in both datasets rather than an architectural limitation. These findings indicate that pairing subword-aware embeddings and informal-text normalization with a soft-voting ensemble of complementary deep learning models is effective for emotion classification in noisy, user-generated social media text, while pointing to class imbalance as the main remaining bottleneck for rare-emotion detection.
Keywords:	Emotion Identification, Social Media Data, Word Embeddings, Deep Learning, Ensemble Classifier
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	A QUANTUM EXPLAINABLE DEEP LEARNING FRAMEWORK FOR EARLY DETECTION OF RF COMPONENT AGING IN COMMUNICATION SYSTEMS
Author:	B NARASIMHA SWAMY , CHETHAN L , BURADA SOUJANIA , GEESALA VEERAPANDU , P. KIRAN KUMAR , ANUSHA GUDISE , PATURI GANGA BHAVANI , SARAVANA KUMAR MADAPPA
Abstract:	In modern communication systems, such as 5G networks, satellite communications, and upcoming 6G networks, early detection of radio frequency (RF) component aging is essential for guaranteeing the reliability, availability, and efficiency of the systems. Prediction models using existing approaches are often not very good at modelling complex nonlinear degradation data and are not frequently interpretable, limiting their practical use in predictive maintenance applications. In order to overcome these issues, a novel Quantum Explainable Deep Learning Framework (QXDL-RFAge) is proposed in this paper for early RF component aging detection. The proposed framework combines Adaptive Quantum Feature Encoding (AQFE), a hybrid Quantum Convolutional Neural Network-Bidirectional Long Short-Term Memory (QCNN-BiLSTM), with Spectral-Temporal Dual Attention (STDA), a hybrid approach, and explainable artificial intelligence (XAI) based on SHAP. The pre-processing of the measurements taken by the RF involves normalization and denoising using wavelets. Then, multi-domain features are extracted, and finally, the quantum states are encoded. Then, QCNN and BiLSTM layers are used to learn the spatial and temporal degradation properties, and the dual attention mechanism is employed for selecting the most important aging signatures. The experimental results on the Accelerated RF Component Aging Dataset (ARFCAD) showed that the proposed framework had an accuracy of 98.76%, precision of 98.21%, recall of 98.54%, an F1-score of 98.37%, a specificity of 98.69%, and an average AUC of 0.992, which outperformed the existing machine learning and deep learning methods. Additionally, it was found that frequency drift and gain degradation are the most prevalent measures of RF component degradation when using the SHAP analysis. The results of the experiments validate the proposed QXDL-RFAge framework as an accurate, interpretable, and reliable predictive maintenance solution that enhances the resilience, sustainability, and operational efficiency of next-generation communication infrastructures.
Keywords:	Quantum Deep Learning; RF Component Aging; Explainable Artificial Intelligence (XAI); Predictive Maintenance; Quantum Convolutional Neural Network (QCNN); Communication Systems.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	FACTORS AFFECTING AI TECHNOLOGY ON EXPORT PERFORMANCE OF CHINESE CROSS-BORDER E-COMMERCE ENTERPRISES
Author:	XIAO -TING LIN , MOHAMMAD NAZIR AHMAD , MARA RIDHUAN CHE ABDUL RAHMAN
Abstract:	Drawing on the TOE framework, this study examined how AI technology affected the export performance of Chinese cross-border e-commerce (CBEC) enterprises through four factors including insufficient technological adaptability, insufficient AI integration in supply chain management (SCM), talent gap, and market competition pressure, with the first three identified as AI implementation barriers and the last as a positive competitive driver. Multiple linear regression and confirmatory factor analysis were performed on 469 valid survey responses, using SPSS 29.0 and AMOS 24.0. The results showed that insufficient technological adaptability (β = -0.248, p < 0.001), insufficient AI integration in SCM (β = -0.239, p < 0.001), and talent gap (β = -0.333, p < 0.001) negatively affected export performance, while market competition pressure had a positive effect (β = 0.224, p < 0.001). All four hypotheses were supported, with the talent gap showing the strongest negative coefficient. This study confirmed the TOE framework's applicability in CBEC, identifies the talent gap as a more critical bottleneck than technological infrastructure, and reveals that competitive pressure positively drives export performance in fragmented international markets, contrasting with domestic e-commerce findings. The results offer strategic guidance for CBEC enterprises to prioritize AI talent development, modular system integration, and leveraging competition for innovation.
Keywords:	AI Technology, Cross-border E-commerce, Export Performance, Influencing Factors, TOE Theory
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	HAM-NET: HIERARCHICAL ATTENTION-BASED MULTI-SCALE NEURAL NETWORK FOR ROBUST IMAGE CLASSIFICATION
Author:	BLESSY MADDA, DR. N. KRISHNAVENI, KUGAN A/L POONGAVANAM, GOTTALA PARAMESWARA KUMAR, KOTNANI PVSR VINAY KUMAR, DR. DEEPAK DHARRAO, S. CHANDRAKALA, DR. CHUNDURI KIRAN KUMAR
Abstract:	Deep learning has fundamentally transformed computer vision, enabling remarkable advances in image recognition, object detection, semantic segmentation, and generative modelling. Convolutional Neural Networks (CNNs) have served as dominant backbones leveraging local receptive fields to capture spatial hierarchies, while Vision Transformers (ViTs) have demonstrated the power of global self-attention for long-range contextual modelling. Despite these complementary strengths, no single architecture has fully bridged the gap between efficient local texture extraction and scalable global context reasoning — a limitation that directly impedes performance on real-world visual tasks requiring simultaneous awareness of fine-grained details and scene-level semantics. The central research problem this work addresses is: how can a unified deep learning architecture simultaneously capture multi-scale feature representations across fine, mid, and coarse granularities, selectively attend to discriminative spatial and channel-wise information at each scale, and adaptively fuse these representations without incurring the quadratic computational complexity of full self-attention or the receptive-field rigidity of standard CNNs? Existing approaches — including Feature Pyramid Networks (FPNs) [10], SE-blocks [14], CBAM [15], and Swin Transformers [7] — address individual facets of this problem but not all simultaneously. To resolve this gap, this paper proposes HAM-Net (Hierarchical Attention-based Multi-scale Network), a novel architecture comprising three synergistic modules: (i) a Multi-Scale Feature Extractor (MSFE) with parallel 3×3, 5×5, and 7×7 convolutional branches capturing complementary feature granularities; (ii) a Hierarchical Attention Module (HAM) applying dual-branch channel and spatial attention gates in parallel to each scale; and (iii) an Adaptive Fusion Layer (AFL) that learns context-dependent scale importance weights via a softmax-gated global context vector. HAM-Net achieves Top-1 accuracies of 97.6% on CIFAR-10, 84.3% on ImageNet-1K, and 76.1% mAP on MS-COCO, outperforming state-of-the-art models including Swin-T [7] and EfficientNet-B7 [5] by up to 5.4 percentage points, while reducing inference latency by 23% and operating with only 18.7M parameters.
Keywords:	Deep Learning, Hierarchical Attention, Multi-Scale Feature Extraction, Image Classification, Vision Transformer, CNN, Feature Fusion, Scale Consistency Regularization, Transfer Learning, Low-Data Regime
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	MAKING SYSTEMS WORK: ALIGNING ACCOUNTING PRACTICES WITH CLOUD ERP DURING IMPLEMENTATION
Author:	YUYUN AUSTY , ROSALINE TANDIONO
Abstract:	Cloud Enterprise Resource Planning (ERP) implementation extends beyond the technical deployment of a digital platform. Enterprise systems embed standardized assumptions concerning data structures, business processes, organizational roles, and control mechanisms. This study examined how alignment between organizational accounting practices and Cloud ERP system logic was achieved during implementation. To answer this question, the study adopted a qualitative single-case study of Enterprise Institution, a pseudonym for a private higher education institution implementing a Cloud ERP system. Data were collected through semi-structured interviews with fourteen participants, comprising three management representatives, four project management team members, and seven key users. The final analysis was organized around three related findings: the new alignment demands created by Cloud ERP implementation, the activities through which workable alignment was developed, and the organizational conditions that enabled those activities. The findings show that alignment did not result from technical implementation alone or from complete organizational conformity to standardized system arrangements. Instead, it developed as organizational actors and consultants clarified requirements, evaluated system possibilities, negotiated alternative responses, and continued to refine organizational and system arrangements during and after implementation. The resulting alignment was therefore workable and revisable rather than perfect or final. This study contributes to the ERP and accounting information systems literature in two ways. First, it develops an empirically grounded account of how workable alignment between organizational accounting practices and Cloud ERP system logic is progressively achieved during implementation. Second, it shows how Cloud ERP alignment reshapes accounting responsibilities, control practices, and cross-functional relationships within an integrated organizational environment.
Keywords:	Cloud ERP, Organizational Accounting Practices, System Logic, Alignment, Enterprise Systems
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	PAF-NET: AN END-TO-END PEOPLE ANALYTICS FRAMEWORK INTEGRATING MULTI-MODAL BIOMETRIC IDENTIFICATION FOR CLOSED-AREA VIDEO SURVEILLANCE
Author:	MARWA EL-NASHAR, MOHAMED ROHAIM1, KHAIRY ASSAR, ALY M. EL-SEMARY
Abstract:	Automated peoples identification in institutional environments is challenging when facial information is unavailable or unreliable. This paper presents the People Analytics Framework (PAF-Net), a multi-camera system that combines person detection and tracking, face detection and verification, and person re-identification. The system uses facial evidence when a usable face is available. When facial information is unavailable, the system uses person re-identification to identify the individual based on appearance. The framework was evaluated using a locally collected campus dataset from five recording sessions and four cameras. The dataset contains 671 person tracks and 24,721 person-crop images. In total, 1,695 identification queries were evaluated. A usable face was available in 287 queries, while the remaining 1,408 queries were processed using person Re-ID. The system correctly identified 280 face-based queries and 1,393 Re-ID queries, giving an overall identification accuracy of 98.7%.
Keywords:	Face Detection, Face Verification, Person Re-Identification, Surveillance System, Deep Learning
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	INTELLIGENT CLOUD-ENABLED CARDIOVASCULAR RISK STRATIFICATION USING HYBRID BIDIRECTIONAL LSTM WITH MULTI-SCALE TEMPORAL ATTENTION AND QUANTUM-INSPIRED HYPERPARAMETER OPTIMIZATION
Author:	GANNI APARNA , DR. HARI JYOTHULA
Abstract:	Cardiovascular diseases (CVDs) remain the leading global cause of mortality, accounting for over 17.9 million deaths annually according to the World Health Organization. Traditional risk assessment methods often fail to capture the complex temporal dependencies and non-linear relationships inherent in cardiovascular physiological data, limiting their predictive accuracy in clinical settings. This study introduces a novel intelligent cloud-enabled framework that integrates a hybrid Bidirectional Long Short-Term Memory (BiLSTM) network with multi-scale temporal attention mechanisms and quantum-inspired hyperparameter optimization for enhanced cardiovascular risk stratification. The proposed architecture employs a three-tier approach: (1) a data preprocessing pipeline incorporating advanced feature engineering with wavelet decomposition and statistical moment extraction; (2) a hybrid BiLSTM model enhanced with multi-head self-attention and residual connections for capturing both short-term fluctuations and long-term trends in cardiovascular biomarkers; and (3) a quantum-inspired optimization algorithm for automated hyperparameter tuning that reduces computational overhead by 67% compared to traditional grid search methods. The model processes multi-modal data including ECG signals, blood pressure variability, heart rate dynamics, and clinical biomarkers through a unified temporal framework. Experimental validation on three benchmark datasets (UCI Heart Disease, MIMIC-III Clinical Database, and PTB-XL ECG Database) demonstrates an average accuracy of 84.26%, sensitivity of 83.36%, specificity of 84.75%, and F1-score of 83.89%, outperforming the strongest baseline on each dataset by 0.80, 2.12, and 3.21 percentage points respectively. The cloud-based deployment architecture ensures scalability with an average response time of 147 ms for real-time risk assessment supporting more than 10,000 concurrent users. Clinical validation across five hospitals with 2,847 patients showed a 23% improvement in early CVD detection rates and a 31% reduction in false positive predictions.
Keywords:	Cardiovascular Disease Prediction, Multi-Scale Temporal Attention, Quantum-Inspired Optimization, Real-Time Analytics, Deep Learning, Healthcare IoT
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	HYBRID QUANTUM PHYSICS INFORMED NEURAL NETWORK FOR TRANSIENT HEAT TRANSFER ANALYSIS IN MICROCHANNEL HEAT SINKS
Author:	NUKALA SUJATA GUPTA , P. ANUSHA , V N V L S SWATHI , SHAHEDA NILOUFER , ANTONY MUTHARASAN S , R.VAISHNAVA KUMAR , H. HUSHNUD , RANJITH KUMAR CHINNAM
Abstract:	The need for thermal control and management systems is critical for sustaining the reliable operation and performance of high-power electronic components under changing thermal environments. In this research, a Hybrid Quantum Physics-Informed Neural Network (HQPINN) model was introduced for effective transient heat transfer analysis of microchannel heat sinks using quantum feature encoding in a physics-informed neural network. The suggested model uses transient thermal information extracted from Computational Fluid Dynamics (CFD) simulation, quantum feature mapping, and a composite physics-constrained loss function consisting of governing heat conduction equation, heat convection equation, boundary, and initial conditions to achieve physically consistent predictions. The performance of the HQPINN model was analyzed based on Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), coefficient of determination (R²), partial differential equation (PDE) residual error, and computational efficiency. As a result of experiments, the suggested HQPINN model demonstrated the performance of 0.62 K MAE, 1.03 K RMSE, 1.58% MAPE, 0.997 R² score, and 0.009 PDE residual error, surpassing the performance of traditional ANN, CNN, LSTM, Quantum Neural Network (QNN), and standard Physics-Informed Neural Network (PINN) in terms of accuracy while decreasing the inference time to 3.9 ms per sample. Integration of quantum feature mapping increased nonlinear thermal feature representation capabilities, while embedding physical constraints improved convergence stability and prediction consistency. These findings demonstrate the potential of the HQPINN model as an efficient computational framework for real-time transient thermal analysis and intelligent cooling system optimization in next-generation electronic and microelectronic applications.
Keywords:	Hybrid Quantum Physics-Informed Neural Network (HQPINN); Microchannel Heat Sink; Transient Heat Transfer; Quantum Feature Encoding; Physics-Informed Neural Networks; Thermal Management.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	QUANTUM GRAPH NEURAL NETWORK FOR PREDICTIVE MAINTENANCE OF INDUSTRIAL GEARBOXES USING VIBRATION AND ACOUSTIC EMISSION SIGNALS
Author:	H. HUSHNUD , N. UMASANKARI , SRINIVASA RAO PALLAPU , TADI SATYA KUMARI , K DEVIPRIYA , P. ANUSHA , B. YELLAMANDA RAO , N. SRIJA
Abstract:	The gearbox systems are vital in manufacturing processes because any failure results in production downtime and increased maintenance cost. In this research work, a new method using Quantum Graph Neural Network (QGNN) has been proposed for the prediction of maintenance requirements of industrial gearboxes using vibration and acoustic emission signal fusion. The first step includes signal processing and converting the signals into multidomain feature representation. The feature vectors are treated as graph nodes with their correlations converted into graph edges. Finally, the Quantum Graph Neural Network (QGNN), using graph convolution and graph attention techniques, is used to learn the complicated spatial relationships and quantum-enhanced feature representations to efficiently classify gearbox fault types. The proposed algorithm is tested on the PHM 2009 Gearbox Challenge dataset and achieves an accuracy of 98.72%, precision of 98.41%, recall of 98.23%, F1-score of 98.32%, specificity of 99.04%, and Area Under the Curve (AUC) of 99.11%. It can be concluded that the proposed technique offers effective, reliable, and robust gearbox fault diagnosis with improved computational efficiency. The developed QGNN framework in this research work is expected to facilitate intelligent predictive maintenance of gearboxes by reducing unexpected equipment failures, minimizing maintenance costs, improving operational reliability, and contributing towards Industry 4.0-based smart manufacturing systems.
Keywords:	Quantum Graph Neural Network (QGNN); Predictive Maintenance; Gearbox Fault Diagnosis; Vibration Signal Analysis; Acoustic Emission; Multi-Sensor Fusion.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	BLOCKCHAIN-ENABLED FUZZY MACHINE LEARNING FRAMEWORK FOR CYBERSECURITY IN CLOUD CONNECTED ELECTRIC VEHICLES
Author:	J RAVINDRA BABU, RAYUDU SRINIVAS , S MANI KUCHIBHATLA , A PRANEETHA , M SIVASANKARA RAO, A V KRISHNARAO PADYALA, D N V SATYANARAYANA
Abstract:	The increasing connectivity among electric vehicles (EVs), charging infrastructure, in-vehicle networks, and inter-vehicle communication has created new cybersecurity challenges for intelligent transportation and smart-grid environments. This study presents a secure and intelligent cybersecurity framework for EV systems by integrating blockchain technology, cloud-based smart-grid communication, and fuzzy machine learning. A smart-grid-assisted cloud computing architecture is employed to collect, process, and exchange EV-related information, while a Fuzzy Adversarial Q-Stochastic (FAQS) model is developed to identify and assess potentially malicious activities. To further strengthen information protection, role-based access control is incorporated to regulate data access according to user responsibilities and authorization levels. Cryptographic mechanisms are applied during data encryption and decryption to protect sensitive EV information against unauthorized access and manipulation. The proposed framework is experimentally evaluated using representative cybersecurity datasets under multiple performance indicators, including security rate, root mean square error (RMSE), quality of service (QoS), scalability, and energy efficiency. The experimental results demonstrate that the proposed approach achieves approximately 98.4% energy efficiency, 96.8% QoS, 92.3% scalability, and 95.7% security rate, indicating its potential to provide secure, efficient, and scalable communication for interconnected electric vehicle and smart-grid environments.
Keywords:	Electric Vehicles, Cybersecurity, Blockchain, Cloud Computing, Fuzzy Machine Learning, Fuzzy Adversarial Q-Stochastic Model, Smart Grid, Role-Based Access Control, Data Encryption, Intelligent Transportation Systems.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	ADAPTIVE CLIMATE AWARE DEEP LEARNING FOR MULTI SOURCE SATELLITE DATA FUSION AND SPATIO TEMPORAL CLIMATE PREDICTION
Author:	MANOJ KUMAR VEMULA , A. VINEELA , P AMBA BHAVANI , D. NAGA PURNIMA , MANI MOHAN DUPATY , R RAJARAMESH MERUGU , P. ANUSHA , ANNAVARAPU MAHALAKSHMI
Abstract:	Predicting climate precisely from heterogeneous satellite and atmospheric data is a difficult task due to the differences in spatial resolutions, temporal characteristics, and varying importance of variables. An adaptive climate-aware deep learning (ACADL) framework was introduced to accurately predict climate using multi-source MODIS, Sentinel-2, ERA5, and OCO-2 data. The introduced ACADL framework includes CNN-based spatial feature extraction, Bi-LSTM-based temporal modeling, climate-context encoding, variable and temporal attention mechanisms, event-aware weighting, and adaptive feature fusion. Experiments have shown the accuracy to be 97.4%, where RMSE is 0.15, MAE is 0.10, and R^{2} is 0.96, while at the same time maintaining a prediction latency of 102 ms. Comparison with ARIMA, Random Forest, CNN, LSTM, CNN-LSTM, and DL-STF frameworks indicated better prediction performance of the proposed approach. Ablation studies have validated the contribution of climate-context encoding, attention mechanisms, and adaptive fusion in the proposed framework. The ACADL framework has proven to be an effective and computationally efficient approach for multi-source climate prediction and can support real-time environmental monitoring, climate assessment, and early-warning applications.
Keywords:	Climate Prediction; Deep Learning; Multi-Source Data Fusion; Spatio-Temporal Learning; Adaptive Attention; Satellite Data.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	RELIABILITY AWARE MULTIMODAL AI SYSTEM FOR REAL-TIME SMART CLASSROOM ATTENDANCE MONITORING
Author:	OMAR IMAMVERDIYEV , JAMAL TEYMUROV , SABINA MAHMUDOVA , NAZILE MAHMUDOVA , AYISHA AHMADOVA , AFET KARIMOVA , TARANA NURUBEYLI
Abstract:	Accurate and scalable attendance monitoring is a practical challenge in digitally supported higher education, particularly in large classrooms where manual registration is time-consuming, error-prone, and vulnerable to proxy attendance. This study proposes a reliability-aware multimodal artificial intelligence system for real-time smart classroom attendance monitoring. The proposed framework integrates visual recognition, audio-based participation analysis, and text/database-level identity verification through a probabilistic fusion mechanism. The system employs lightweight and edge-deployable models, including YOLOv5 for student detection, MobileNet for face recognition, DistilBERT for identity matching, and CNN-LSTM or wav2vec 2.0 for audio-based activity analysis. To improve decision reliability under non-ideal classroom conditions, the fusion layer dynamically adjusts modality contributions according to the quality of visual, audio, and database evidence. In addition, uncertain cases are redirected to an abstention or manual-review state instead of being automatically accepted. A pilot deployment scenario was evaluated in a smart classroom at Mingachevir State University using recognition accuracy, inference latency, multimodal consistency, and computational efficiency as the main performance criteria. The integrated system achieved an average attendance recognition accuracy of 93.2 ± 0.7% with a decision latency of 45–55 ms, demonstrating its suitability for real-time classroom operation. The results indicate that reliability-aware multimodal fusion provides more stable attendance decisions than single-modality configurations and can reduce the risk of incorrect registration under occlusion, illumination variability, background noise, and incomplete identity evidence. The proposed system provides a practical basis for reliable AI-supported attendance monitoring in smart educational environments.
Keywords:	Multimodal Artificial Intelligence; Smart Classroom; Attendance Monitoring; Edge AI; Reliability-Aware Fusion; Real-Time Recognition
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	A MULTI-MODAL SELF-SUPERVISED PREPROCESSING FOR BONE TUMOR SEGMENTATION AND CLASSIFICATION IN MEDICAL IMAGING
Author:	HIMABINDU GOLI , Dr. N. SUPRIYA
Abstract:	Accurate tumor detection and segmentation of bone from medical imaging modalities, such as X-rays and magnetic resonance imaging (MRI), remain challenging tasks owing to limited annotated datasets, inter-institutional data variability, and the complexity of anatomical structures. This study presents a novel preprocessing-driven self-supervised learning framework called M3-BonePrep (Multi-modal multi-Institutional bone preprocessing), which aims to improve representation learning and diagnostic performance in bone tumor analysis. The framework has five main modules: cross-modality intensity harmonization (CMIH), adaptive bone structure localization (ABSL), noise-aware radiological enhancement (NARE), topology-preserving data augmentation (TPDA), and self-supervised data quality scoring (SS-DQS). This study introduces a novel domain-aware preprocessing framework, M³-BonePrep, which combines four aspects of intensity harmonization, anatomical bone localization, noise-aware enhancement, topology-preserving augmentation, and self-supervised data quality scoring into a single pipeline. It is now known that, by using domain-specific, quality-aware pre-processing, the adverse impact of modality variation, noise, structural inconsistencies, and limited annotations can be mitigated, leading to improved self-supervised representation learning for heterogeneous bone tumor images. The proposed approach was evaluated on benchmark datasets, including BTXRD (X-ray) and MSTT-199 (MRI), for the classification and segmentation tasks. The experimental results demonstrate that the proposed framework achieves better performance than baseline methods, with a classification accuracy of 96.85% and segmentation Dice scores of 0.89 (BTXRD) and 0.86 (MSTT-199). Furthermore, the model has an AUC of 0.99, which is very good discrimination capability. Qualitative and quantitative analyses confirm better delineation and robustness of the boundaries across imaging modalities. The proposed framework provides a scalable and data-efficient solution for bone tumor analysis and can therefore be deployed in real-world, resource-constrained clinical environments.
Keywords:	Bone Tumor Analysis, Self-Supervised Learning, Medical Image Preprocessing, Multi-Modal Imaging (X-Ray, MRI), Deep Learning Segmentation.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	QUANTUM MACHINE LEARNING FRAMEWORK FOR PREDICTIVE YIELD ENHANCEMENT IN NANOSCALE VLSI MANUFACTURING
Author:	S.SAI KUMAR , SUDHEER NIDAMANURI , P SUSHMA , T VEERRAJU , KAJA KRISHNA MOHAN , TATARAO VANA , SK. MABUNNI , P. ANUSHA
Abstract:	The Nanoscale Very Large-Scale Integration (VLSI) manufacturing for achieving high production yield has been seriously complicated by various factors: process variation, lithographic errors, equipment-induced variation, and increasingly complex fabrication procedures. In this study, the authors present a novel Quantum Machine Learning Framework for Predictive Yield Enhancement (QML-PYE) that can be used to accurately predict wafer yield and optimize semiconductor manufacturing processes. The proposed framework includes a Quantum Feature Selection Module (QFSM), a Variational Quantum Yield Predictor (VQYP), and a Quantum Bayesian Process Optimizer (QBPO) to simultaneously optimize the process, predict the yield, and select the critical features of the process. The SECOM semiconductor manufacturing data set was employed for the experiments, consisting of 1,567 wafers and 590 process parameters. The framework uses quantum state encoding and variational quantum circuits to encode complex nonlinear relationships between manufacturing variables. The experimental results showed that the proposed QML-PYE framework has an accuracy of 98.12%, precision of 97.86%, recall of 97.41%, F1-score of 97.63%, and RMSE of 0.053, which exceeds the accuracy, precision, recall, F1-score, and RMSE of conventional machine learning and deep learning models. Furthermore, the manufacturing yield was improved as a result of the proposed optimization strategy from 88.60% to 95.40%, with a 7.68% yield enhancement. The results show that quantum-enhanced learning can bring great improvements to predictive accuracy and the ability to optimize the workflow in semiconductor manufacturing. The proposed framework is effective and suitable for intelligent manufacturing solutions for the next generation of nanoscale VLSI fabrication, which would further lead to better productivity, lower manufacturing costs, and autonomous smart semiconductor factories.
Keywords:	Quantum Machine Learning; Nanoscale VLSI Manufacturing; Yield Prediction; Variational Quantum Circuits; Semiconductor Process Optimization; Wafer Yield Enhancement.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	HYBRID QUANTUM GRAPH NEURAL NETWORK FOR MULTIMODAL ALZHEIMER’S DISEASE PROGRESSION PREDICTION
Author:	P. ANUSHA , YAVARNA NAGAPRAMODKUMAR , NARASIMHA RAO THOTA , D.V.L.PRASANNA , SRILAKSHMI RAMYA SAKAMUDI , PENDEM SANTHI , R. SARAVANAN, NIRMAL MAHESH
Abstract:	Predictive diagnosis of early-stage AD disease progression is essential to enable timely clinical assessment of patients. However, the existing deep learning algorithms usually use each imaging modality independently, whereas conventional graph-based learning algorithms are unable to effectively capture multimodal heterogeneity and complex nonlinear relationships between brain regions and clinical features. In order to tackle this problem, this research paper presents a Hybrid Quantum Graph Neural Network (HQGNN) for predicting the progression of Alzheimer's disease using multimodal data. The proposed model makes use of structural Magnetic Resonance Imaging (MRI), FDG Positron Emission Tomography (PET), cognitive assessment scores, and demographic characteristics from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. The brain Regions of Interest (ROIs) are modeled using graph nodes, and the Graph Attention Network (GAT) model is utilized to learn relationships among different brain regions. Then, Variational Quantum Circuits (VQCs) are utilized to augment the nonlinear feature representation ability of the neural network, followed by multimodal fusion of features and classification using a Softmax-based classifier into five AD disease stages: Normal Control (NC), Early Mild Cognitive Impairment (EMCI), Late Mild Cognitive Impairment (LMCI), Mild Cognitive Impairment (MCI), and Alzheimer's Disease (AD). The proposed HQGNN achieved an accuracy of 98.82%, precision of 98.41%, recall of 98.26%, F1-score of 98.33%, specificity of 99.12%, and AUC of 99.03%, outperforming the evaluated models, including CNN, ResNet50, Vision Transformer, Graph Convolutional Network, and Graph Attention Network. The findings provide evidence that integrating multimodal neuroimaging with graph-based relational learning and quantum-enhanced feature representation can produce a unified representation for distinguishing multiple stages of AD. The study therefore contributes a hybrid quantum–graph learning framework and demonstrates its potential for multimodal disease progression prediction.
Keywords:	Alzheimers disease, Hybrid Quantum Graph Neural Network (HQGNN), Multimodal Neuroimaging, Graph Attention Network, Variational Quantum Circuit, Early Disease Prediction
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	A SEMANTIC AND CONTEXT AWARE NLP FRAMEWORK FOR AUTOMATED TEXT ERROR DETECTION
Author:	ENAMALA.KALAVATHI , DR.E.S.SHARMILA SIGAMANY , G.USHASRI , BASSA DEVI , P MADHAVI, BHUVANESWARI PAGIDIPATI, NAGULMEERA SAYYED, P. ANUSHA
Abstract:	Non-native English writers face various problems related to errors at grammatical, lexical, punctuation, and contextual levels that can negatively impact academic and professional writing. In this paper, a new framework for automatic writing error detection and explanation is suggested, called Context-Aware Neural Error Detection and Explanation (CANE). It uses contextual language processing techniques for detecting, categorizing, and explaining various types of writing errors. The proposed framework combines text preprocessing, Bidirectional Encoder Representations from Transformers (BERT)-based contextual embeddings, dependency parsing, semantic consistency analysis, and multi-class classification of writing errors at grammatical, lexical, and contextual levels. The major research contribution is the integration of contextual semantics, dependency relations, semantic consistency, and explanation generation into one framework for writing error detection. The proposed framework is evaluated on the benchmark BEA-2019 Shared Task and First Certificate in English (FCE) datasets by Accuracy, Precision, Recall, F1-score, Matthews Correlation Coefficient (MCC), and Area Under the Curve (AUC). CANE framework achieves Accuracy 98.91%, Precision 98.63%, Recall 98.47%, F1-score 98.55%, MCC 97.92%, and AUC 99.14%. The obtained results are higher compared to other rule-based, recurrent neural network, and transformer-based models considered in this study. Integration of contextual embeddings, grammatical dependency analysis, and semantic reasoning allows detecting context-dependent errors and decreasing false positive detections. Additionally, the explanation generation component gives users informative feedback, corrective suggestions, and interpretable recommendations about how to fix their mistakes. Overall, the proposed framework demonstrates the potential of integrated contextual and explainable linguistic analysis for automated writing assistance in educational platforms, language-learning applications, and academic writing support for non-native English learners.
Keywords:	Context-aware Natural Language Processing (NLP); Writing Error Detection; Grammatical Error Detection; BERT; Explainable Artificial Intelligence (XAI); Non-Native English Writing.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	DIGITAL TECHNOLOGIES FOR NOTARIZATION OF FAMILY LAW AGREEMENTS AND THEIR INTEGRATION WITH ELECTRONIC DOCUMENT MANAGEMENT SYSTEMS
Author:	TETІANA ANDRUSHCHENKO , LIUDMYLA PYSKUN , NATALIIA BORTNIK , MARYNA TOPORKOVA , ANTON DEMCHUK
Abstract:	The study addressed the need for indicator-verified integration of digital notarization with electronic document management (EDM) to ensure legal admissibility and process stability. It aimed to develop and empirically verify a techno-legal framework for this integration. The methodology combined ETL data preparation, process mining, quasi-experimental modelling, SLA verification, and MCDA synthesis. This sequence linked reproducible provenance measurements to a formal assessment of legal admissibility and interoperability. The analysis identified stable integration scenarios, causal effects, and a minimally sufficient admissibility profile. Legal stability depended on integration quality rather than digital form alone. Scenarios S2–S3 accumulated 60–75% of process defects, whereas the STP-oriented S1 scenario reduced notarization time by 18–35%, integration delay by 22–40%, and formal defects by 10–25%. The findings showed that scenario interoperability and provenance completeness jointly determine the legal robustness of digitally notarized family-law agreements. The resulting indicator-verified techno-legal framework translates these relationships into reproducible admissibility criteria. Future research should test the framework across other contract categories, jurisdictions, and long-term judicial outcomes.
Keywords:	Family Law Relations, Family Law Agreement, Notarization, Digital Notary, Electronic Document Flow, Legal Interoperability, Civil Process
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	DUAL-ATTENTION FUSION NETWORK FOR BEARING FAULT DIAGNOSIS USING HYBRID DEEP LEARNING WITH VIBRATION SIGNAL ANALYSIS
Author:	DR. SUDHAKAR JYOTHULA , MADHAVI PERLA , DR. KATTA NAGARAJU , T. KRISHNA MOHANA , DR. KISHORE KUMAR K , BABYMOL KURIAN , CHALLAPALLI SUJANA , DR. VADDEMPUDI SUJATHA LAKSHMI
Abstract:	Bearing failures in rotating machinery systems are a major cause of unplanned downtime and equipment failure. Traditional machine learning techniques suffer from the limitations of manual feature vectorization and low generalization ability in diverse working scenarios. We design a new network architecture, Dual-Attention Fusion Network (DAFN), that consists of three parallel branches — 1D-CNN, BiLSTM and Transformer encoder — fused through a novel dual-level attention module composed of channel attention (CAM) and spatial attention (SAM) designed to recalibrate and fuse multi-scale features learned directly from raw vibration signals in an adaptive way. The proposed model is tested on the Case Western Reserve University (CWRU) and MFPT bearing datasets under diverse noise environments with SNRs from −10 dB to +20 dB. Experimental results show that DAFN achieves 96.7% classification accuracy and AUC of 0.985 on the CWRU benchmark dataset, outperforming standalone CNN, LSTM and Transformer models by margins of 9.4%, 7.1% and 5.5% respectively. The model sustains an F1-score of 74.2% at −10 dB compared to 61.3% for the best competing method. Ablation studies confirm the dual attention module alone provides a 3.2% cumulative accuracy gain. Without architectural modification, DAFN achieves 95.3% accuracy on MFPT, with inference latency of 9.3 ms meeting real-time industrial requirements.
Keywords:	Attention Mechanism, Bearing Fault Diagnosis, Bidirectional LSTM, Deep Learning, Dual-Attention Fusion, Transformer Encoder, Vibration Signal Analysis
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	HYBRID MULTIPLE-MODE FUSION ARCHITECTURE FOR DETECTING LONG-RANGE RELATIONSHIPS AND FINE-GRAINED STRUCTURAL IMAGE FEATURES
Author:	KAMARSU APPARAO , PAVAN KUMAR PAGADALA , BUDATI ANIL KUMAR ,AKELLA RAMAKRISHNA
Abstract:	Traditional multiple-modal medical imaging systems suffer from poor contrast, irregular tumor borders, and noise artifacts, whereas modern image processing algorithms for multiple-modal medical imaging are important for accurately segmenting lung tumors. Differences in cancer shape and strength between CT, PET-CT, and MRI lead to different interpretations and delays in clinical decisions. The following are the obstacles that the proposed study aims to tackle: timely diagnosis, improved radiation planning and accurate response monitoring to treatment. This article proposes a hybrid multiple-mode fusion architecture based on a transformer-improved UNet, 3D nnU-Net, and diffusion-mediated refinement for detecting long-range relationships and fine-grained structural features. High-performance methods like hierarchical concentration (Swin-UNet), dual-stream feature arrangement (multi-modal), and self-supervised contrast learning are employed to reduce the need for large labeled datasets. Adaptive attention fusion improves the PET metabolic map by integrating with CT anatomical accuracy, while boundary-aware loss functions enable detection of small lesions. Experimental results on public datasets (including the LIDC-IDRI and PET-CT databases) demonstrate higher segmentation accuracy, up to 4-6% higher Dice coefficient than the standard convolutional neural network (CNN) models, and a lower number of false positives. The results show significant stability improvements through multi-modal deep fusion, and the clinical implications highlight its potential for real-time, repeatable, and highly accurate lung tumor classification.
Keywords:	Image Processing, Lung Tumors, Unet, 3D Nnu-Net, Swin-Unet, Multi-Modal, Segmentation, And Classification
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	QUANTUM ENHANCED RF CIRCUIT PARAMETER OPTIMIZATION FOR LOW POWER IOT TRANSCEIVERS
Author:	V N V L S SWATHI , ANJANEYULU NELLURU , PADMAJA PRIYADARSINI P , G L NARASAMBA VANGURI , N. NITHYA LAKSHMI , SUNEETHA BANDEELA , NAVYA DEEPTHI TUMMALA , VASANTHI YARRA
Abstract:	The number of low-power Internet of Things (IoT) devices is increasing, and efficient radio frequency (RF) circuit optimization techniques are becoming increasingly attractive because of their potential for enhancing the performance of communication systems without consuming too much energy. The authors introduce a Quantum-Enhanced RF Circuit Parameter Optimization (QERPO) framework in this paper, which integrates deep surrogate learning with quantum-inspired optimization to optimize the parameters of a low-power IoT transceiver. First, 10,000 circuit configurations are created for a simulated RF circuit. The configurations are then used as inputs to a deep surrogate learning model that is able to accurately predict various RF performance metrics with fewer electromagnetic simulation cases. A quantum-inspired optimization engine is then used to optimize several objectives simultaneously: minimizing power consumption, minimizing noise figure, maximizing gain, maximizing bandwidth, maximizing return loss, and maximizing power-added efficiency. It is experimentally proven that the proposed framework achieves better performance in terms of power consumption (13.4 mW), gain (20.3 dB), noise figure (1.48 dB), bandwidth (142 MHz), power-added efficiency (39.8%), and optimization error (0.045%) when compared with other conventional optimization methods like GA, PSO, DE, and BO. Furthermore, the proposed approach is faster than the existing optimization approaches as it includes surrogate learning and quantum-inspired search in the convergence process. Based on these results, it is concluded that the QERPO framework is an efficient, scalable, and computationally effective solution for designing energy-efficient RF transceivers, making it highly suitable for the next generation of IoT and future wireless communication systems.
Keywords:	Quantum-Inspired Optimization, RF Circuit Parameter Optimization, Low-Power IoT Transceivers, Deep Surrogate Learning, Multi-Objective Optimization, Energy-Efficient Wireless Communication.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	A QUANTUM FEDERATED LEARNING FRAMEWORK FOR PRIVACY PRESERVING BRAIN TUMOR CLASSIFICATION ACROSS DISTRIBUTED MEDICAL CENTERS
Author:	G. LALITHA KUMARI, A. JAFFAR SADIQ ALI , M. LENIN KUMAR , V. LAKSHMI SAILAJA , GUMMEDELLI SRISHAILAM , MD. AAMIR SOHAIL , RAMAKRISHNA PRADHANA , T. ASWINI DEVI
Abstract:	Classification of brain tumor types based on Magnetic Resonance Imaging (MRI) images plays a crucial role in accurate diagnosis and developing appropriate strategies for treatment. Centralized Deep Learning models suffer from several shortcomings such as data privacy, security, and data-governance issues associated with transferring sensitive medical images from patients among different medical centers. For this purpose, a novel Quantum Federated Learning Framework (QFLF) is proposed for brain tumor classification of medical images of Glioma, Meningioma, Pituitary Tumor, and Normal Brain classes in this paper without sharing raw data from different institutions or hospitals. The proposed scheme utilizes the advantages of Quantum Computing and Federated Learning to enable privacy-preserving collaborative learning across distributed medical centers. In the suggested scheme, MRI images pass through stages such as CNN-based feature extraction, Quantum Feature Encoding, hybrid quantum-classical federated learning, Adaptive Weighted Federated Learning (AWFL), and secure exchange of encrypted model parameters. The performance evaluation of proposed QFLF was done by taking BraTS 2021 Dataset containing four classes of brain tumors. The proposed method attained accuracy, precision, recall, F1-score, and specificity values equal to 99.12%, 98.91%, 98.74%, 98.82%, and 99.28%, respectively. It has outperformed other conventional CNN models such as CNN, ResNet50, DenseNet121, and Vision Transformer. Additionally, federated training accuracy improved significantly from 90.25% to 99.12%, and federated training loss reduced from 0.452 to 0.082, indicating stable convergence during federated training. These findings demonstrate that the proposed framework can combine privacy-preserving collaborative learning with quantum-enhanced feature representation for accurate brain tumor classification. However, validation was limited to a simulated federated environment and the BraTS 2021 dataset. Future work should evaluate the framework using heterogeneous multi-center clinical datasets and practical quantum computing environments.
Keywords:	Quantum Federated Learning; Brain Tumor Classification; Magnetic Resonance Imaging (MRI); Privacy-Preserving Learning; Quantum Machine Learning; Distributed Medical Centers.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	INTEGRATION OF QUANTUM COMPUTING AND DEEP LEARNING: CHALLENGES AND PROSPECTS
Author:	RAMYA CHITTIMALLA , P.SUDHEER KUMAR, K. SATHISH , K. ANU RADHA, SUBBA RAO POLAMURI, HARI JYOTHULA , KROSURI SIVA PRASADA RAO, SAMATHA CHOWDARY PONDURI
Abstract:	Deep learning and quantum computing have each individually demonstrated transformative potential, yet no established framework exists that effectively bridges these two paradigms for practical tasks such as fault detection in power systems — a gap this work directly addresses. This paper presents a novel hybrid framework that integrates deep learning and quantum computation for improved anomaly detection in electric power systems. The proposed approach extends quantum-contrastive learning training methods to cover computational tasks involving deep and conditional restricted Boltzmann machines. The framework demonstrates a significant leap in speedup and accuracy, with notably reduced training time, by combining TensorFlow-based machine learning with a Quantum-Deep Learning (QDL) Interface on a state-of-the-art quantum processor. Experimental evaluations on image recognition and natural language processing tasks demonstrate that the QDL approach achieves 94% and 89% accuracy respectively, outperforming classical deep learning methods by 7 and 8 percentage points. Training time is reduced by 30% and a quantum speedup factor of 10× is observed. Statistical analysis via t-tests (p = 0.017 for image recognition; p = 0.042 for NLP) confirms the significance of these improvements at a 95% confidence interval. The paper further discusses challenges including technology readiness, algorithmic complexity, and quantum-classical system integration, alongside substantial opportunities such as quantum advantage, enhanced model optimization, and hybrid computing architectures. The work represents a critical step toward realizing the combined potential of quantum computing and deep learning for next-generation computational intelligence. The findings confirm that hybrid quantum-classical architectures are not only technically feasible but deliver statistically validated performance advantages, positioning quantum-deep learning as a viable direction for next-generation computational intelligence.
Keywords:	Quantum-Deep Learning Interface, Hybrid Framework, Computational Intelligence, Quantum Computing, Deep Learning, Uncertainty Quantification, Variational Quantum Circuit
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	MLNETXAI-FUSION: A HYBRID INTELLIGENT ANALYTICS SELF-HEALING FRAMEWORK FOR MACHINE LEARNING BASED DIAGNOSIS AND MITIGATION OF NETWORK ISSUES IN WIRELESS NETWORKS
Author:	PEDIREDLA SRILATHA, Dr. SUNEEL PAPPALA, Dr. SRIPADA RAMA SREE
Abstract:	As the number of IoT devices and wireless distributed environments increases, the wireless communication systems are more susceptible to network problems, such as congestion, adverse traffic flow, network route instability, packet loss, decreasing latency, and hacking. These conventional network monitoring tools lack adaptivity, scalability, and statical methods, making it difficult to support adaptively diagnose and autonomously mitigate network issues. In this paper, the authors have come up with a new Hybrid Intelligent Analytics Self-Healing Framework for Machine Learning Based Diagnosis and Mitigation of Network Issues in Wireless Networks (MLNetXAI-Fusion) to solve the aforementioned problems. These are integrated into an intelligent architecture: Adaptive Feature Engineering, Deep Behavioural Anomaly Mining, Hybrid Ensemble Learning, Graph-Based Contextual Intelligence, Explainable Analytics, Trust Aware Optimization and Reinforcement Learning Self-Healing Strategies. The learning latent traffic behavior is achieved with deep autoencoder, and in the second step, the abnormal network activities are detected by conducting the reconstruction analysis. Finally, a hybrid XGBoost-LightGBM ensemble model is employed to effectively detect network problems and the communication relationship between the nodes and the propagation of anomalies in context are modeled using a Graph Attention Network. The SHAP and LIME modules are Explanatory Analytical Modules which are used to improve Interpretability and Trustworthy Decision Analysis. Furthermore, an adaptive self-healing engine estimates the attack level, isolates suspicious nodes, optimizes the routing behavior and optimizes the Quality of Service (QoS) parameters with the reinforcement-based routing optimization algorithm. The experimental performances have been tested with the reference test datasets of wireless network intrusion, and the result shows that the performance of the three modules: anomaly detection, trust stability, QoS optimization and autonomous recovery, is better than the conventional methods. The proposed framework gives an intelligent, wireless network management and network autonomous recovery deployment and scaling solution.
Keywords:	Hybrid Intelligent Analytics, Autonomous Network Management, Explainable Analytics, Trust-Aware Networking, Network Anomaly Detection, QoS Optimization, Wireless Networks, Self-Healing Networks, Graph Attention Network, Reinforcement Learning.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	MULTI-LEVEL REPRESENTATION LEARNING FOR INTRACRANIAL HEMORRHAGE DETECTION USING DEEP CONVOLUTIONAL NETWORKS
Author:	Dr. LAKSHMI NAGA JAYAPRADA GAVARRAJU , DR. GADALA SUHASINI , DR.MOHAMMED SIRAJUDDIN , KASICHAINULA VYDEHI , K. MALARKODI , VIJAY KUMAR LAKSHETTY , M PALLAVI , Dr. D.BHAVANA
Abstract:	Intracranial Hemorrhage (ICH) is a serious cerebrovascular disease that involves bleeding in the cranium, and timely diagnosis is crucial to the outcome of treatment. Computer-aided neuroimaging analysis has great potential to assist in a timely clinical decision-making process, but current CNN models might not be able to learn the heterogeneous anatomical, geometric and textual features of intracranial bleeding. To address this challenge, a dual-network deep learning approach based on RepVGG and VGG19 is designed for automatic identification of ICH in CT and MRI examinations. The proposed pipeline involves image-intensity standardization and comprehensive data augmentation to enhance data diversity and model robustness, followed by representation learning using transfer-learning. Maps derived from different levels of hierarchy are fused with each other in a complementary manner from two pretrained networks, yielding a map containing both global anatomical features and local hemorrhagic features. The integrated features are then fed to a series of optimally connected layers with dropout to promote stability in the classification and prevent overfitting. When assessed with 4000 carefully selected neuroimages, performance was reported as 97.2% accuracy, 96.8% sensitivity, 96.6% specificity, 96.0% F1-score and an area under the receiver operating characteristic curve (AUC) of 0.975. Based on the experimental results, the integrated RepVGG + VGG19 is shown to be more effective in classification than standalone RepVGG and VGG19 and some of the latest deep learning methods. The results also indicate that a hierarchical feature integration approach can yield a more informative representation of subtle hemorrhagic manifestations and that the proposed approach can be a computer-assisted tool for rapid and scalable ICH assessment.
Keywords:	Intracranial Hemorrhage (ICH), Advanced VGG Models, Feature Fusion, Neuroimaging, Computer-Aided Diagnosis, Medical Image Analysis
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	GIS-BASED SHRIMP FISHERY SUPPLY CHAIN SYSTEM MODEL USING GETIS-ORD GI* AND ANT COLONY OPTIMIZATION ALGORITHMS
Author:	NURDIN , TAUFIQ, FAJRIANA, RAHMAD
Abstract:	Shrimp fisheries production is one of the leading commodities of the aquaculture sector in the North Aceh coastal region of Indonesia, which plays a vital role in supporting the economy of coastal communities. Information on production distribution patterns and supply chain efficiency is still limited, requiring spatial-based analysis to support decision-making. This study aims to identify areas of concentrated shrimp production using the Getis-Ord Gi* method and optimize the supply chain using the Ant Colony Optimization algorithm. This study was conducted in 22 sub-districts spread across Bireuen, North Aceh, and East Aceh Regencies with a total Vannamei Shrimp production of 6,231.29 tons per year. The Getis-Ord Gi* analysis results indicate that Nurussalam District was identified as a hotspot with a GiZScore of 2.2946 and a p-value of 0.0218, while Lapang and Baktiya Barat Districts were identified as coldspots with a GiZScore of -2.3479 and a p-value of 0.0189. The remaining 19 sub-districts were in the not-significant category. Furthermore, supply chain optimization was carried out on 61 ponds connected to three main storage points, namely Peudada Fish Landing Base, Krueng Mane Fish Landing Base and Idi Fishery Port. The results of the Ant Colony Optimization algorithm application showed that the total distribution distance could be reduced from 1905.92 km to 1905.60 km, resulting in a distance saving of 0.32 km with an efficiency level of 0.02%. In addition, Peudada District had the lowest distribution cost value of 1.39 km, while Nurussalam District as a hotspot area had a distribution cost value of 16.07 km. The results of the study showed that the integration of the Getis-Ord Gi* method and the Ant Colony Optimization algorithm based on the Geographic Information System was able to identify priority production areas and produce more efficient distribution channels to support the development of the Vannamei Shrimp supply chain in the North Coast region of Aceh, Indonesia.
Keywords:	Ant Colony Optimization (ACO), Getis-Ord Gi*, Supply Chain, Shrimp Fisheries, Geographic Information System.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	AN INTELLIGENT ADAPTIVE HYBRID ENCRYPTION FRAMEWORK FOR HEALTHCARE CLOUD SECURITY VIA MACHINE LEARNING-BASED DECISION OPTIMIZATION
Author:	Mrs. P. R. SUKANYA SRIDEVI, Dr. R. SHANTHI, Dr. V. HEMA, Dr. A. CATHREEN GRACIAMARY
Abstract:	In recent years, cloud computing has become a key platform for the management of significant amounts of healthcare data; however, cloud technology has also created serious challenges in keeping the data secure and keeping the patient’s privacy intact in fast changing and limited resource situations. Traditional cryptography techniques (i.e., AES, RSA, Blowfish, ABE and Homomorphic Encryption) provide strong guarantees for securing information but serve mainly as static methods with little or no ability to adapt to varying types of data and changing environmental/system conditions. Given these issues, this study proposes an AI based Adaptive Secure Hybrid Encryption (ASHE), which combines a machine learning driven decision model to allow for context-aware and dynamically driven decision making on the types of cryptographic techniques to use. More specifically, a supervised learning method using a Random Forest classifier is used to determine optimal encryption methods based on multidimensional input characteristics such as the size of data, type of data, sensitivity of data, privacy requirements and workload of the system. The proposed technique implements a hybrid cryptographic framework that combines AES for fast and efficient data encryption, RSA for secure key exchange, and SHA-3 for maintaining data integrity. By integrating these three mechanisms, the framework establishes a strong and comprehensive security architecture capable of protecting sensitive healthcare information in cloud-based environments. To evaluate the effectiveness of the framework, a healthcare dataset containing 500 diverse records was developed. The dataset included multiple forms of healthcare information such as electronic health records (EHRs), medical imaging files, and several system-level performance attributes designed to simulate realistic cloud computing conditions. The proposed framework was then tested and compared with conventional encryption approaches and advanced cryptographic models using different performance metrics, including encryption latency, throughput, CPU utilization, and overall processing efficiency. The experimental findings revealed that the machine learning-enabled Aggregate Secure Healthcare Environment (ASHE) framework achieves a balanced trade-off between computational performance and security robustness. In addition, the framework supports privacy-preserving and responsive data protection techniques that are suitable for modern healthcare systems. The study further highlights how the integration of machine learning with hybrid cryptographic principles can contribute to the development of scalable, intelligent, and resilient security solutions for next-generation healthcare cloud infrastructures.
Keywords:	ASHE,SHA, Encryption, Cloud Healthcare, ML, Decision Support
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	FROM RANKING TO FEASIBLE ASSIGNMENT: A TRACEABLE NUMERICAL AUDIT FOR TEACHER PLACEMENT
Author:	ROSHAYATI OSMAN, MOHD ZAKREE AHMAD NAZRI , AZLINDA ARSHAD1, AZLIN RAMLI
Abstract:	Teacher placement becomes a joint allocation problem when several candidates compete for limited vacancies. A ranked list alone cannot ensure that the final set of placements is feasible or explain why a highly ranked pair was not selected. This deterministic design-science study tests a traceable score-to- assignment audit protocol using an archived AHP weight vector, a five-by-six benefit matrix, TOPSIS ranking, binary rules, complete enumeration and a structured explanation record. The coded teacher-school pairs form a constructed numerical instance; no population sampling or inferential test is used. TOPSIS recalculation produced coefficients from 0.3399 to 0.6792. The base model had 13 feasible solutions among 32 combinations and an optimum of 1.8112. A competitive vacancy-closure scenario had three feasible two-placement solutions; the optimum was 1.1957. A stricter exact-placement scenario had one feasible solution, assigning G2 to S1 and moving G1 to S2 despite the higher local score of G1-S1. The active capacity and placement rules explain this change. The novelty is procedural rather than algorithmic: suitability, joint feasibility, exact verification and contrastive explanation are connected in one inspectable evidence chain. The evidence supports numerical auditability for the three stated scenarios, not field effectiveness or large-scale performance.
Keywords:	Teacher Placement, Decision Support System, TOPSIS, Binary Optimisation, Constraint Audit, Counterfactual Explanation
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	VARIATIONAL QUANTUM NEURAL NETWORKS FOR EARLY SKIN CANCER DIAGNOSIS
Author:	SRINIVAS MULKALAPALLI , Y SUREKHA , VASAVI BANDE , BASSA DEVI , AFREEN FATIMA MOHAMMED , PUNYALA RAMADEVI , P SAI ARUN MIHIR , ARUNA VIPPARLA
Abstract:	Skin lesion classification is a challenging task owing to intra-class variations and visual similarities between different lesions. Although conventional deep-learning methods have achieved promising results, the application of compact variational quantum models for multiclass skin lesion classification remains insufficiently explored. In this study, we introduce a Variational Quantum Neural Network (VQNN) combined with a lightweight CNN for multiclass skin lesion classification using the ISIC 2019 dataset. The novelty of the proposed approach lies in combining classical CNN-based spatial feature extraction with four-dimensional feature compression and a four-qubit variational quantum circuit with ring-based entanglement. The extracted features are encoded into four qubits and processed through four variational quantum layers before classification. The proposed method achieved an accuracy of 96.82%, precision of 96.51%, recall of 95.94%, F1-score of 96.21%, specificity of 99.31%, and AUC of 98.17%. Among the eight lesion classes, melanocytic nevus achieved the highest F1-score of 98.56%, while dermatofibroma showed comparatively lower performance. Ablation analysis demonstrated progressive improvement from the classical CNN baseline to the complete four-layer VQNN. The findings provide new evidence that compact variational quantum processing combined with classical feature extraction can achieve competitive multiclass skin lesion classification while using only four quantum qubits, establishing a computationally compact hybrid framework for further investigation in medical image analysis.
Keywords:	Skin Lesion Classification; Variational Quantum Neural Network; Quantum Machine Learning; Deep Learning; ISIC 2019; Medical Image Classification
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	MULTIMODAL DEEP LEARNING FOR ANDROID MALWARE DETECTION: COMBINING STATIC CODE, DYNAMIC BEHAVIOR, AND NETWORK TRAFFIC
Author:	Dr G.SREENIVASULA REDDY, Dr.CH.SITA KAMESWARI, Dr. K. PRABHAVATHI, JAKKAPU NAGALAKSHMIDEVI, J S V GOPALA KRISHNA, SUGANYA BABY.N, PANIBHATE NEELAKANTESWARA, Dr.T. KANIMOZHI, K.MOHANA LAKSHMI, M PRAVEEN KUMAR, Dr.T.VENGATESH
Abstract:	The rapid proliferation of Android malware has rendered traditional signature-based detection methods increasingly ineffective against sophisticated, obfuscated threats. While deep learning has emerged as a powerful paradigm for malware detection, most existing solutions rely on a single modality either static code analysis, dynamic behavioral analysis, or network traffic analysis leaving them vulnerable to the inherent limitations of that particular analysis type. Static analysis alone is susceptible to code obfuscation, packing, and dynamic code loading; dynamic analysis faces incomplete code coverage and conditional trigger evasion; and network analysis can be circumvented through encryption and traffic mimicry. Although recent multimodal approaches have attempted to combine complementary information sources, they predominantly focus on two modalities and lack a comprehensive integration of all three feature types. Furthermore, existing hybrid frameworks often employ either early or late fusion in isolation, failing to leverage the complementary strengths of both strategies. This paper addresses these gaps by proposing a multimodal deep learning framework that integrates static, dynamic, and network traffic features to achieve robust Android malware detection. Our architecture employs a hybrid fusion mechanism combining Convolutional Neural Networks (CNNs) for static code analysis, Long Short-Term Memory (LSTM) networks for dynamic behavioral sequence modeling, and Graph Neural Networks (GNNs) for network traffic pattern analysis. We evaluate our framework on the CICAndMal2017 and CICInvesAndMal2019 datasets, achieving 98.3% accuracy with a false positive rate of 1.8%, outperforming unimodal approaches by 6–11%. The experimental results demonstrate that multimodal fusion significantly enhances detection performance and resilience against obfuscation techniques, providing a promising direction for next-generation mobile security solutions.
Keywords:	Android Malware Detection, Multimodal Deep Learning, Feature Fusion, Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), Graph Neural Networks (GNN), Dynamic Behavior Analysis, Network Traffic Analysis, CICAndMal2017, Obfuscation Resilience.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	CAMPUSPULSE AI: INTELLIGENT AI-POWERED FRAMEWORK SMART CAMPUS FOR WI-FI OPTIMIZATION
Author:	PREETHA S, MAHALAKSHMI B S , U G GAGAN SAI , VIKRAM R VANJRE , VINEET KULKARNI , YUVARAJ A R
Abstract:	Managing Wi-Fi networks on large university campuses is a real and complex challenge. Take for instance, multiple people and devices connected with varying network traffic while staying mobile around a vast campus vicinity. Traditional solutions for constant and consistent data connectivity have always been a challenge in today's smart campuses. Hence, CampusPulse AI, a smart system that uses AI to make campus Wi-Fi function smoothly and efficiently. CampusPulse AI gathers network info in real-time, uses a smart model to spot congestion, predicts network load, and then suggests fixes autonomously. At any given point of time, the model investigates a minimum of 14 parameters for each Wi-Fi access point (AP) to determine the signal strength, traffic and the end points. The processing is designed to execute in under 200 milliseconds to ensure a faster resolution. The device was tried and tested with 120 APs and the results were as expected. The overall network speed increased by 41.2%, packet loss dropped by 52.4%, and congestion was reduced by 38.6%. With a congestion spotting accuracy of 96.3%, the device did yield higher results in comparison to traditional devices. The proposed system is being designed to manage up to a maximum of 1,000 APs without any losses in the network speed making it suitable for smart campus deployments and future intelligent wireless networks.
Keywords:	Smart Campus, Wi-Fi Optimization, Artificial Intelligence, Network Congestion Detection, AI-Driven Network Management;
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	MECHANISM FOR TRANSFER OF INDUSTRIAL DIGITAL TWINS TO SMART CITY INFRASTRUCTURE
Author:	OLHA PROKOPENKO , TALIAT BIELIALOV , VITALIY OMELYANENKO , MILENA KIRILOVA FILIPOVA , JULIA ROSEND
Abstract:	Introduction: Digital twins are increasingly used to manage complex urban infrastructure. At the same time, the transfer of industrial models to smart city environments remains poorly formalized. This necessitates verification of the semantic-adaptive approach to their deployment. Purpose: The aim of the study is the laboratory-based verification of the preservation of states, functions, fault logic, KPIs, and decision-support values during the deployment of an industrial digital twin in an urban cyber-physical environment. Methods: The methodology was based on a reproducible laboratory verification of the transfer between a digital twin of a MathWorks Simscape Triplex Pump pumping unit and a digital twin of a DHALSIM urban water supply network. Structural and parametric analysis determined the suitability of parameters, fault classes, and functions for urban deployment. Comparative model mapping compared direct transfer with functional transformation of fault logic. Semantic-adaptive deployment provided semantic adaptation of parameters, states, functions and KPIs of the pump digital twin to the urban infrastructure environment. Multi-criteria evaluation and statistical testing compared the digital twin transfer mechanisms (M1, M2 and M3) in terms of transfer quality. The analytical environment was implemented in Python using pandas, NumPy, SciPy, statsmodels, and matplotlib. Results. Transfer quality depended on preserving model logic rather than only transferring parameters. IUDTTI increased from 0.46 for M1 Direct Parameter Transfer to 0.88 for M3 Semantic-Adaptive Model Deployment. ANOVA showed a statistically significant difference between the mechanisms: F(2,12) = 38.6; p < 0.001; η² = 0.86. These findings support the laboratory advantage of semantic-adaptive transfer over direct parametric transfer. Practical reuse of industrial twin modules remains subject to field validation. Conclusion: The scientific contribution is a semantic-adaptive transfer mechanism that extends transfer beyond parameters to state models, function models, event logic, and KPI semantics. Future work requires field testing on real urban sites, heterogeneous assets, and more complex operating scenarios.
Keywords:	Semantic Compatibility, Cyber-Physical Environment, Risk Scenarios, Functional Integrity, Maintenance, Hydraulic Simulation, Interoperability
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	AN INTELLIGENT SEQUENTIAL NAVIGATION FRAMEWORK FOR OBSTACLE PROGRESSION ANALYSIS AND RISK-AWARE ASSISTIVE GUIDANCE FOR VISUALLY IMPAIRED INDIVIDUALS
Author:	D. KARTHIKA , S.P. BALAMURUGAN
Abstract:	Traditional assistive navigation systems for visually impaired people evolved early pieces of research on the understanding of sequential obstacles, adaptation of movements to adjust risk levels and reasoning about intelligent navigation. To overcome these limitations, this paper present an intelligent assistive navigation framework called SNAP-RAG (Sequential Navigation and Progression-based Risk-Aware Guidance) that integrates lightweight obstacle perception, sequential obstacle progression analysis, adaptive risk scoring, context-aware navigation decision generation and voice-guided assistance into a single real-time framework. First, in light of the computationally efficient obstacle detection under various navigation environments, a Lightweight Spatial Attention Feature Boosting (LSAFB) model based on YOLOv5n is designed. Then, Sequential Obstacle Progression Module (SOPM) examines spatial obstacle progression in order to estimate relative distance and directional position. A novel Adaptive Navigation Risk Scoring (ANRS) framework is then proposed to assess obstacle severity granularity-wise using spatial, positional, and progression-aware contextual features. Also, a Context-Aware Navigation Decision Engine (CNDE) combines obstacle risk and environmental context before providing intelligent movement guidance in order to make safe navigation decisions. Lastly, an assistive voice guidance module executes the generated navigation directions as immediate speech responses to engage with users. The experimental results validate that the proposed framework realizes prominent obstacle understanding, navigation robustness as well as assistive guidance performance against state-of-the-art models and with a lightweight computational complexity for CPU-based real-time deployment. The proposed framework provides an efficient and scalable approach for intelligent assistive navigation in the visually impaired environment.
Keywords:	Assistive Navigation System, Sequential Obstacle Progression Risk-Aware Navigation, Obstacle Detection, Context-Aware Guidance, Visually Impaired Assistance, Intelligent Mobility Framework, Voice-Guided Navigation
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	PERFORMANCE ANALYSIS OF INFINITE-FAILURE NHPP SOFTWARE RELIABILITY MODELS USING POWER-TYPE LIFETIME DISTRIBUTIONS
Author:	SEUNG KYU PARK
Abstract:	This study analyzed and evaluated the performance of an infinite-failure NHPP reliability model by applying Power-type lifetime distributions (Weibull, Extended-Weibull, Rayleigh, Weibull-Extension) that can reflect the characteristics of failure rates changing in the form of powers of time and are suitable for various reliability change patterns. For this study, software failure times were utilized, and parameter calculation of the presented model was solved using the MLE procedure. By utilizing various performance analysis methods (selection of the best-fitting model using MSE and R², estimation power for the true value using m(t), analysis of instantaneous failure intensity using λ(t), and evaluation of future reliability using R̂(τ)), it was confirmed that the Weibull-Extension model was the most efficient among the competing models. Therefore, this study newly elucidated the reliability performance of Power-type distribution models, which had been lacking in previous research. Furthermore, it is expected that if software developers efficiently utilize these research results, they can use them as basic testing data necessary for failure rate analysis during the early stages of development. These results could serve as useful information for meaningful design guidelines and testing.
Keywords:	Extended-Weibull, Infinite-Failure NHPP, Power-type, Rayleigh, Weibull, Weibull-Extension.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	ADDRESSING DATA SKEWNESS AND FIT ISSUES IN REACTIVE CLOUD SCALING: THE BAYESOPT-SMOT-SVM ALGORITHM APPROACH
Author:	SATYA NAGAMANI POTHU , DR. SWATHI KAILASAM
Abstract:	This research addresses critical challenges in reactive auto-scaling within cloud computing's Infrastructure as a Service (IaaS) framework, focusing on data skewness and model underfitting/overfitting issues that impede workload prediction accuracy. The study involves a comprehensive evaluation of ten prominent workload prediction models applied to datasets from leading cloud service providers: Azure, AWS, and Google Cloud. It uncovers a significant shortfall in these models: their limited ability to handle data skewness and their vulnerability to underfitting and overfitting, resulting in reduced prediction efficacy. To overcome these obstacles, the research introduces the BayesOpt-SMOT-SVM algorithm, a hybrid methodology combining Bayesian Optimization (BayesOpt), Synthetic Minority Over-sampling Technique (SMOT), and Support Vector Machine (SVM). This algorithm is innovatively designed to optimize the predictive performance by effectively addressing the identified issues. Extensive simulations reveal the BayesOpt-SMOT-SVM algorithm's enhanced capabilities in comparison to existing models, showcasing its improved accuracy and scalability for workload prediction in reactive cloud scaling scenarios. While highlighting its strengths, the paper also discusses the limitations of the proposed algorithm, thus paving the way for future research. This contribution enriches the cloud computing field by providing a robust solution to enhance workload prediction accuracy in IaaS environments.
Keywords:	Bayesian Optimization, Data Skewness, Underfitting, Overfitting, Reactive Cloud Scaling, Workload Prediction, IaaS, BayesOpt-SMOT-SVM Algorithm, Machine Learning, Cloud Computing.
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text

Title:	DEVELOPMENT AND VALIDATION OF THE CARE BRIDGE MODEL: A DIGITAL HEALTH-SYSTEM COORDINATION PLATFORM FOR ELDERLY CARE SERVICES
Author:	MOHD RASIDI SARBINI, MOHD FAHMI MOHAMAD AMRAN, NOR ASIAKIN ABDULLAH, SURESH THANAKODI, SHARIFAH AISHAH SYED ALI, RITA WONG MEE MEE, LIM SEONG PEK, RASITHRA RAVICHANDRAN, MUHAMMAD FAIRUZ ABD RAUF
Abstract:	As global populations age, it creates significant and increasing demand for efficient and accessible ways of supporting elderly in becoming and being independent. Current elderly care systems can be described as fragmented, reactive, and lacking properly organized methods of establishing connection between users and relevant services. The Care Bridge Model is presented in this research as a method to overcome fragmentation in elderly care systems. Care Bridge is a digital health coordination platform that helps patients to effectively navigate among the medical services that provide rehabilitation, care, and community support. The proposed model consists of three domains, which are User Domain, Service Provider Domain, and Support Network Domain. They are united by the deterministic matching engine that enables systematic connections of users and services based on location, availability, and cost. Design Science Research was used in the study. Development process included problem identification, artefact design, prototyping, demonstration, and evaluation of the prototype. Evaluation of the prototype included expert review with the involvement of three experts and user acceptance testing with twenty participants. Thematic analysis was used in identifying the six main themes which emerged from qualitative feedback and quantitative evaluation with the use of Likert scale. Hence, this study findings indicate that criteria-based matching has potential to support service relevance, accessibility, and user experience in elderly care coordination, while expert and user evaluations provide preliminary evidence of the model’s practical feasibility.
Keywords:	Elderly Care, Deterministic Matching, Mobile Application, Service Coordination, Digital Health
DOI:
Source:	Journal of Theoretical and Applied Information Technology 15^th October 2026 -- Vol. 104. No. 19-- 2026
Full Text