		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
July 2025 | Vol. 103 No.14

Title:	HYBRID OPTIMIZATION TECHNIQUES FOR MOBILITY-AWARE, ENERGY-EFFICIENT SMALL CELL DEPLOYMENT IN 5G NETWORKS
Author:	S. VINODH KUMAR, A. VIJAYALAKSHMI, A. PACKIALATHA, B. EBENEZER ABISHEK
Abstract:	Expanding wireless communication networks is necessary to meet the growing number of mobile devices and the demand for faster internet. One practical way to increase network capacity and coverage in heavily populated regions is to deploy tiny cells. Smaller cells require more energy, increasing operating costs and negatively affecting the environment. Traditional deployment approaches ignore user mobility, despite its substantial impact on network performance. We present a strategy for microcell deployment in 5G networks, utilizing hybrid optimization techniques to address issues related to mobility awareness and energy efficiency. The planned teenTo improve data transfer capacity and increase user density in tiny cells, the suggested strategy clusters users using a Modified Smell-Bees Optimization (MSBO) algorithm. This research introduces a Gannet Optimal Induced Cuckoo Search (GOCS) approach to grouping microcells into optimal locations while accounting for various design limitations. This book lays out an Improved Coral Reef Optimization (ICRO) approach that takes reliability criteria into account for better coral reef optimization. Measures such as connection quality, user mobility, congestion rate, and mean time to failure are part of these criteria. Assisting in the setup of compact base stations is the goal of this plan. Simulations conducted in the Google Colab environment greatly enhance important Quality of Service (QoS) measures. The MSBO-GOCS-ICRO model is better than the well-known GSCP, TIPA, and ECM-BPSD models in many ways. For example, it cuts convergence time by 49%, increases the number of small base stations in use by 64%, and makes the network 154% more energy efficient. These findings indicate that the suggested approach is the optimal choice for the deployment of tiny cells in 5G networks.
Keywords:	5G networks, Small cell deployment, Hybrid optimization, Energy efficiency, Mobility management, Quality of Service.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	PREDICTING FRAUD: A MACHINE LEARNING APPROACH TO SECURE TRANSACTIONS IN CREDIT CARD SYSTEM
Author:	BASHAR I. HAMEED, MOHAMMED A. MOHAMMED, HUMAM K. YASEEN
Abstract:	The expansion of e-commerce has uncovered extensive vulnerabilities in web-based transactions, creating opportunities. The enormous use of credit cards in online transactions, motivated by their perks like discounts and bonuses, has resulted in a substantial upward thrust in credit card fraud. Conventional strategies, including hand checks and inspections, even as traditionally employed, have proven to significant obstacles in identifying fraudulent actions due to their time-intensive nature, high cost, and imprecision. The emergence of Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL)-based techniques provides a promising innovative solution for addressing fraudulent activities with the aid of permitting the pattern recognition and anomaly detection of financial transactions. Even with recent advances in research into ML-based credit card fraud detection, the imbalance in credit transaction data makes identifying fraudulent activities a challenging task. This paper presents an advanced credit card fraud detection system using ML and DL algorithms; however, it is very important to investigate the scenario of anomaly detection concerning its characteristics. The paper analyzes a specific case study of the credit card dataset, highlighting the important preliminary steps in creating the necessary processes before the proposed model is applied. Our experimental results demonstrate that the Logistic Regression model achieved superior performance in evaluation metrics compared to the other models tested in our experiment.
Keywords:	Fraudulent Financial Transactions, Credit Card, Fraud Detection, Machine Learning, Deep Learning
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	BIG DATA ADOPTION IN INTERNATIONAL LOGISTICS MANAGEMENT
Author:	HASSAN ALI AL-ABABNEH, OLHA MAIBORODA, IHOR KUKIN, VADYM YAKOVETS, LIUBOV HRYNIV
Abstract:	The Big Data adoption increasingly affects the international logistics management, thereby improving operational efficiency and competitiveness. The aim of this study is to analyse the impact of the use of Big Data (BD) sets on the logistics efficiency of companies in different countries for 2019–2023. The study uses regression modelling and case studies together with econometric analysis of companies to determine how technology infrastructure, regulatory environment, market dynamics, operational performance, and cultural context influence the causal relationship between BD and logistics efficiency. There is limited research on how BD adoption interacts with contextual factors in global logistics environments. The results demonstrate that deeper BD adoption is associated with higher business logistics efficiency, and the effectiveness of this interaction is moderated by the quality of the regulatory framework and technological infrastructure. The study shows the significant influence of cultural factors on organizational approaches to the BD use. The study also emphasizes that companies must invest in new technologies and adapt to regulatory changes to fully realize the benefits of BD opportunities. The findings support the role of strategic management methods that incorporate BD sets into logistics operations to make logistics efficient and innovative. Further research may focus on changing the role of digitalization in logistics and its implications for international supply chain management. It is also appropriate to conduct sector-specific analysis for a deeper understanding by policymakers and practitioners to improve logistics processes through data-driven strategies. The article proposes a new concept of ‘intelligent adaptive logistics (IAL)’ based on the integration of Big Data, artificial intelligence, IoT and cognitive analytics to create self-learning logistics systems. The purpose of the development of ‘intelligent adaptive logistics (IAL)’ is to assess how the level of digital maturity of a company in the field of IAL affects its logistics efficiency in the international context. For the first time, the IAL_Adopt index is proposed, which comprehensively takes into account not only the implementation of Big Data, but also the use of AI, IoT, cognitive analytics and the level of adaptation to changes in the external environment. The results of the panel regression analysis show that a high level of IAL_Adopt significantly improves logistics efficiency, especially in countries with high-quality infrastructure and a favourable regulatory environment. Despite technological investment, companies in weak regulatory and infrastructure contexts cannot fully benefit from BD. The IAL concept and the IAL_Adopt index provide a scalable framework for assessing digital maturity and planning strategic improvements in logistics. These findings confirm the positive impact of integrated digital technologies on logistics performance globally. The developed concept can serve as a basis for digital logistics transformation strategies.
Keywords:	Big Data Sets, Logistics Efficiency, International Management, Econometric Analysis, Technological Infrastructure, Innovation In The Supply Chain, Intelligent Adaptive Logistics, AI, Cognitive Analytics, Logistics Efficiency, Digital Maturity.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	A NOVEL PVDF-COPPER NANOPARTICLE-BASED NANOGENERATOR FOR EFFICIENT ENERGY HARVESTING
Author:	V. VIJAYALAKSHMI, Dr. K. S. GEETHA, Dr. SHANMUKHA NAGARAJ, R. THIAGARAJAN
Abstract:	The speedy development of energy harvesting technologies has highlighted the requirement for efficient and sustainable means of addressing the rising demand for self-powered electronic devices. This research investigates the design of a polyvinylidene fluoride PVDF-Copper nanoparticle-based nanogenerator for energy harvesting purposes, aiming to improve flexibility, biocompatibility, and cost-effectiveness. The simulated nanogenerator is designed and modeled with the help of COMSOL Multiphysics to analyze its physical, electrical, mechanical, and chemical behavior under changing environmental conditions. During the simulation stage, charge buildup, stress distributions, and mechanical deformation are analyzed to optimize the choice of materials and structural design. Once optimized, fabrication is done by employing advanced material deposition and electrode patterning methods to fabricate an experimental prototype with maximum energy conversion efficiency. The artificially designed nanogenerator is thoroughly characterized to evaluate its physical, electrical, chemical, and mechanical properties to determine its reliability and stability in various operational conditions. The characterization involves voltage and current output measurement, mechanical flexibility evaluation, and chemical stability testing by FTIR. Performance evaluation finally takes into account signal conditioning circuits to check the actual practical suitability of the nanogenerator. The experimental results exhibit the efficiency of the nanogenerator in harvesting energy, a peak output voltage of 5.2V, and power density of 120 µW/cm² for an applied force of 10N. Signal-conditioning circuit integration stabilizes the power delivery and renders the nanogenerator very useful in wearables, biomedical implants, and industrial sensors. The findings of this study highlight the potential of PVDF-Copper nanoparticle-based nanogenerators to advance self-powered systems further, paving the way for further research in sustainable energy technologies.
Keywords:	Nanogenerator, Energy Harvesting, PVDF, Copper Nanoparticles, COMSOL Multiphysics, Signal Conditioning Circuits, Mechanical Deformation, Self-Powered Systems, Sustainable Energy Solutions
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	PERSONALISED LEARNING USING DEEP LEARNING TECHNIQUES
Author:	M HARSHITHA SYAMALA, V LIKHITHA, DIVYA LINGINENI, PRASANNA DRL
Abstract:	With the growing popularity of online learning and e-education platforms, the importance for adaptive learning as per individual learning needs has been increasing. Although existing solutions have come a long way, they still lack the ability to capture a wide range of learning styles and truly deliver personalized learning experiences. This project suggests a comprehensive framework combining sentiment analysis and knowledge graph technologies to enhance personalization in educational platforms. With the use of advanced models such as BERT-Bi-LSTM-Attention for analyzing course reviews and knowledge graphs to represent user interaction, course relationships and store user preferences, the system can break cold-start issues and ensure that course recommendations are more transparent. The methodology includes an approach to leverage learner motivation and learning outcomes by delivering a more responsive and personalized digital education environment. The solution also includes a feedback loop and dynamically generates quizzes based on completed courses helping students assess their progress and identify areas for improvement, offering a scalable model that can be utilized to adapt to various educational environments and offers continuous enhancement in personalized learning.
Keywords:	Personalization, Course Recommendations, BERT-Bi-LSTM-Attention, Knowledge Graph, Feedback Mechanism.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	EARLY FIRE HAZARD PREDICTION FRAMEWORK IN SMART CITIES USING DEEP LEARNING WITH ANTLION OPTIMIZATION ALGORITHM
Author:	DR.G. BHUVANESWARI, DR.G. MANIKANDAN, M. SANDHYA, DHANESH KUMAR, DR. ZIAUL HAQUE CHOUDHURY, PRABHAKARA RAO T
Abstract:	This study aims to refine the early fire risk prediction model for evaluating the accurate locations of fire using sensor data. Internet of Things (IoT) is integral to smart cities. IoT applications in smart cities include crime predictions, traffic optimization and monitoring of health and environmental conditions. This article reports a study on using Recurrent Neural Network (RNN) with Ant Lion optimization (ALO) framework to enhance and refine the prediction of fire hazards. IoT sensors in smart cities monitor the environmental conditions such as drought, temperature, smoke, flame, relative humidity, fuel moisture and duff moisture. This sensed data is stored in the firebase cloud storage and analyzed in the MATLAB tool. The ensuing enhancement of the proposed model is validated by comparison with conventional prediction models. Our results indicate gains in accuracy and reduction in error rates in fire hazard predictions.
Keywords:	Environment, fire hazards, Internet of Things, Smart city.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	CYBERSECURITY ASSESSMENT FRAMEWORK FOR HEALTHCARE INSTITUTIONS PRE-MEDICAL CYBER-PHYSICAL SYSTEM ADOPTION
Author:	MOHAMED AWADA ELMAGBARI, MAHEYZAH MD SIRAJ, SITI HAJAR OTHMAN
Abstract:	Medical cyber-physical systems enable the remote monitoring of patients, thereby enhancing accessibility to care. Unfortunately, secure adoption is still difficult, as this remains an unresolved topic that includes cybersecurity and privacy issues. While a number of frameworks exist in the general realm of CPS cybersecurity readiness, few frameworks address the healthcare domain. This study proposes a Cybersecurity Assessment Framework (CAF) together with a quantitative scorecard to assess cybersecurity readiness before healthcare institutions embark on MCPS adoption. Meta-analysis of existing frameworks, coupled with expert interviews, has resulted in five Critical Success Factors (CSFs) being established: reliability, validity, third-party authentication, security, and transparency. Furthermore, a case study approach was adopted with IT managers and healthcare professionals of two Libyan hospitals. Results indicated that the CAF is valid and usable and supports the secure adoption of MCPS, though privacy remains a concern. This work presents a novel, domain-specific CAF for healthcare cybersecurity, followed by tools nurturing IT governance.
Keywords:	Critical Success Factors, Cyber-Physical Systems, Cybersecurity Assessment Framework, Healthcare Industry, Medical Cyber-Physical Systems.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	PREDICTION OF NON-ALCOHOLIC FATTY LIVER DISEASE (NAFLD) USING DNA PATHOLOGICAL DATA AND SUPPORT VECTOR MACHINES
Author:	T. V. K. P. PRASAD, SRI LEKHA BANDLA, N. SRIKANTH, K KAVYA RAMYA SREE, INAKOLLU ASWANI, PAMULA UDAYARAJU, BODDU L V SIVA RAMA KRISHNA
Abstract:	Non-Alcoholic Fatty Liver Disease (NAFLD) has emerged as one of the most prevalent liver disorders globally, affecting nearly one-third of the population, with particularly high incidence rates in countries like the UK. Despite its widespread occurrence, accurate estimation of its prevalence remains a challenge. Early-stage NAFLD, typically characterized by simple steatosis, can silently progress to more severe conditions such as non-alcoholic steatohepatitis (NASH), fibrosis, and cirrhosis if left untreated. This progression significantly compromises liver function and increases the risk of cardiovascular complications. However, current diagnostic methods, including magnetic resonance spectroscopy and ultrasound imaging, are often limited by cost, accessibility, and diagnostic specificity. Given the clinical urgency and the limitations of conventional diagnostics, this study addresses the critical need for an accessible and accurate method to detect early-stage liver disease—specifically, to predict NASH within the NAFLD spectrum. We propose a machine learning-based approach that leverages clinical and pathological data, including blood parameters and ultrasound-derived tissue characteristics, to support early detection. Using a dataset of 181 patients, we applied preprocessing techniques such as normalization and categorical encoding to prepare the data for modelling. Features such as integrated backscatter (IB), Q-factor, and homogeneity factor (HF) were extracted to quantify liver tissue characteristics. Support Vector Machine (SVM), chosen for its balance of simplicity and efficiency in handling high-dimensional datasets, was employed for classification and regression tasks. Experimental validation using Python-based implementations demonstrated the model's effectiveness, achieving an average accuracy of 89.95% across both clinical and imaging-derived datasets. This study underscores the potential of machine learning in improving early diagnosis of liver diseases and reducing their long-term clinical burden.
Keywords:	Fatty Liver Diseases, Non-Alcoholic Fatty Liver Diseases, NASH, SVM, Pathological Information, Machine Learning.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	A NOVEL HYBRID APPROACH FOR BRAIN TUMOR CLASSIFICATION USING LIQUICON-NET AND ENHANCED PREPROCESSING
Author:	PAVAN KUMAR PAGADALA, TAN KUAN TAK, R. THIAGARAJAN, PRAVIN RAMDAS KSHIRSAGAR
Abstract:	This paper introduces a thorough methodology for categorizing brain tumors using the BRATS dataset, utilizing sophisticated image processing and machine learning techniques. The technique commences by obtaining and adjusting brain scan images to a consistent size of 256x256 pixels by spline interpolation. The photos are subsequently transformed into grayscale using the luminosity approach, which simplifies the data while preserving crucial structural details. The Total Variation (TV) denoising technique is utilized to diminish noise and maintain essential characteristics, leading to the production of superior pre-processed photographs. The pre-processed images are subjected to segmentation using the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm, which efficiently classifies data points into core points, border points, and noise. This segmentation process reveals discrete regions of interest within the brain scans. Afterwards, Haralick descriptors are utilized to extract features from the gray-level co-occurrence matrix (GLCM). These features include Contrast, Correlation, Energy, and Homogeneity. These characteristics offer a comprehensive representation of the texture and spatial connections within the photographs. Dimensionality reduction is accomplished by employing t-Distributed Stochastic Neighbor Embedding (t-SNE) to streamline the feature set while maintaining optimal performance. The classification task utilizes the LiquiCon-Net model, which combines Liquid Neural Networks (LNNs) and Convolutional Neural Networks (CNNs). The model integrates ResNet-50's high-level spatial features with the temporal information derived from the Liquid Time-Constant Network (LTCN). The fusion layer combines these features, and the ultimate classification is determined by a sequence of densely connected layers, which are then followed by a softmax function. The performance evaluation of the suggested model is carried out utilizing a confusion matrix, ROC plot, and measures such as accuracy, sensitivity, and specificity. The proposed model attains a precision of 99.42%, a selectivity of 99.62%, and a responsiveness of 99.81%. The results exhibit exceptional performance in comparison to current models, such as EfficientNets, Capsule Networks (CapsNet) + VGG19, and AlexNet + GoogleNet Ensemble. The thorough examination and impressive measurements highlight the efficiency of the suggested approach in precisely categorizing brain tumors, presenting substantial prospects for enhancing diagnostic precision in clinical environments.
Keywords:	Brain Tumors, Glioma, LNN, LTCN, Medical, Deep learning
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	THE USE OF INNOVATIVE TECHNOLOGIES FOR THE PREVENTION AND DETECTION OF CRIMES RELATED TO DOCUMENT FORGERY
Author:	HANNA VALIHURA, IHOR NOVIKOV, MARIIA DIAKUR, OLENA YUSHCHYK, VLADYSLAV KUTSENKO
Abstract:	The relevance of the study is underscored by the proliferation of digital forgery and the need for comprehensive solutions to ensure the authenticity of documents and their legal validity in notarial proceedings. The aim of the study is to substantiate and validate the architecture of preventive forgery detection with increased invariance, compatibility and cryptosecurity. The research employs methods as follows: technological typification, comparative analysis, technical and architectural decomposition, functional enhancement, and simulation prediction. An analysis encompassing 33 services and 7 technology classes revealed that complex SSI solutions exhibit the highest level of completeness. The optimized Microsoft Entra Verified ID demonstrates a reduction of 44.7% in latency, 61.6% in anchoring time, and 57% in cloud usage, alongside a 2.5% enhancement in detection accuracy − thereby substantiating the effectiveness of protocol level forgery resistance. The scientific novelty lies in the development of a SSI architecture that integrates BBS+, ZKP and Edge AI thereby harmonizing crypto resilience, offline validation, and an elevated degree of interoperability. Prospects for future research include the development of a prototype SSI architecture designed for the field testing of efficiency, scalability and resilience under real operational load and variable network conditions, as well as evaluating its suitability for electronic notarial proceedings.
Keywords:	Cryptographic Signatures; Blockchain Anchoring; Decentralized Identifiers (DID); Zero-Knowledge Proofs; Decentralized Storage
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	SECURING 6G WIRELESS TRANSMISSION THROUGH QUANTUM KEY DISTRIBUTION INTEGRATED WITH VISIBLE LIGHT COMMUNICATION
Author:	HARIPRASAD.B,K.P. SRIDHAR
Abstract:	The advent of 6G wireless networks requires strong physical layer security (PLS) measures to guard off ever-changing cyber dangers and guarantee extremely dependable communication. In congested locations, traditional security systems based on radio frequency (RF) encounter challenges such as increased interference, eavesdropping concerns, and spectrum congestion. To tackle these issues, the present research investigates how Visible Light Communication (VLC) might be incorporated into 6G networks to supplement existing methods of improving PLS. Minimizing interference from outside sources during VLC signal transmission, guaranteeing safe key exchange, and finding the optimal location for system performance and security are all significant obstacles. This research presents a new framework Quantum-Enhanced Secure VLC (QES-VLC), which uses adaptive beamforming in conjunction with quantum key distribution (QKD) to improve data security and reduce the probability of interception. This method guarantees a smooth handoff between hybrid VLC-RF networks and restricts illegal access by taking advantage of VLC's inherent directional transmission features. Secure vehicle communications, industrial automation, and smart city infrastructure of the future are all areas where this research could be beneficial. This approach enables ultra-secure, high-speed wireless networks by incorporating VLC into the 6G ecosystem; this improves security without reducing efficiency. An improvement in resilience against eavesdropping is achieved at 99.5%, a decrease in bit error rate to 15.7%, and an increase in secrecy capacity to 98.2%. An extensive simulation analysis is performed using MATLAB to test the suggested approach. The results showed that secure key exchange efficiency improved to 97.3%, while system latency and power consumption are lowered to 28.3% and 18.9%, respectively. In addition, its effectiveness in secure data transfer has been experimentally validated in a real-time VLC experimentation.
Keywords:	6G, Physical Layer, Security, Incorporating, Visible, Light, Communication, Transmission, Quantum, Key Distribution
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	ENHANCING FAULT DIAGNOSIS AND IMPROVING PRODUCTIVITY IN INDUSTRIAL MANUFACTURING USING DEEP LEARNING TECHNIQUES
Author:	RAJESH KUMAR VERMA, KALLI SRINIVASA NAGESWARA PRASAD, K.S.RANJITH, V.S.N.MURTHY4,I.NAGA PADMAJA, P.THIRUMOORTHY,BH.KRISHNA MOHAN
Abstract:	Fault diagnosis in industrial manufacturing is a critical issue that affects the productivity and efficiency of manufacturing processes. Outdated methods for fault diagnosis often rely on manual inspections, which are time-consuming and prone to errors. This framework proposes a deep learning-based fault diagnosis model to improve productivity in industrial manufacturing. The JAYA optimization algorithm and Fast Grid Search (FGS) are employed to optimize the hyperparameters of the model. The proposed model is implemented in MATLAB software and evaluated using a dataset of industrial manufacturing process data. The results show that the proposed model achieves high accuracy and precision in fault diagnosis, outperforming traditional methods. The model can identify faults early, reducing downtime and improving overall productivity. The findings indicate that predictive maintenance and optimized feature importance significantly enhance performance metrics and reduce downtime, with notable improvements in accuracy up to 0.99 and substantial cost savings, contributing to a return on investment of around 85%. The development of a more efficient and reliable fault diagnosis system for industrial manufacturing has contributed to this framework. Future scope includes integrating the proposed model with other machine learning algorithms and incorporating sensor data from multiple sources to further improve its performance.
Keywords:	Deep Learning, Fault Diagnosis, Industrial Manufacturing, JAYA Optimization Algorithm, Fast Grid Search, and CNN Architecture.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	SECURE SOFTWARE DEVELOPMENT THROUGH AI-DRIVEN CODE VULNERABILITY ANALYSIS
Author:	RAMAKRISHNA KOLIKIPOGU, NEHA BELWAL, DR.SABITHA KUMARI FRANCIS, DR. S. N. V. JYOTSNA DEVI KOSURU, DR SUBBA RAO POLAMURI, RAMESH BABU PITTALA
Abstract:	Software development has been drastically transformed by the emergence of automatic vulnerability detection systems generated by the rapid expansion of artificial intelligence (AI). Application programs frequently encounter security threats due to error-prone & time-consuming traditional techniques of vulnerable code detection. Using ML and DL methods, the present study proposes an AI-driven solution to code vulnerability detection. Using Natural language processing (NLP) and analysis of static code, the recommended approach identifies possible risks, including buffer overflow, SQL injection, as well as cross-site scripting (XSS). To optimise code security with the aid of the framework's inbuilt AI-based detection of vulnerabilities and real-time feedback. The outcomes of the performance assessment show that the claimed AI-based model excels traditional methods in terms of both error identification accuracy and efficiency. This research illustrates the significant role of AI in enhancing software security and gives useful insights into forthcoming developments in cybersecurity driven by AI.
Keywords:	Software Development, Machine Learning, Deep Learning, Cybersecurity, Natural Language Processing, Vulnerable Code Detection.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	BLOCKCHAIN - ENABLED DIGITAL IDENTITY MANAGEMENT FOR ENHANCED SAFETY SYSTEMS
Author:	Dr D Naga Tej, Dr R V V Murali Krishna, Dr Chinnarao kurangi, Sagar Sathuluri, Dr Subba Rao Polamuri, Dr Hari Jyothula
Abstract:	Digital identity management that incorporates blockchain technology provides a safe and effective way to enhance safety systems. Typical issues with older methods of managing identities include data breaches, identity theft, and unauthorized access. This encrypted, decentralized, and irreversible blockchain method ensures 100% accurate identification verification by eliminating any weak points and consolidating power. By further automating the authentication process, smart contracts aim to boost transparency while lowering fraud risks. This paper focuses on how safety-critical settings may leverage blockchain-based digital identity management systems to provide data accessibility, integrity, and privacy. In addition to assisting corporations in meeting and exceeding security laws, blockchain, a distributed ledger technology, gives individuals more control over their data. The suggested method discourses problems like interoperability, scalability and regulatory constraints, paving the way for a more secure and reliable structure. The primary emphasis of this research is on the revolutionary potential of blockchain technology as it pertains to identity management. Therefore, current safety systems will have enhanced dependability, security and efficiency.
Keywords:	Blockchain, Data Breaches, Digital Identity Management, Distributed Ledger, Safety Settings.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	ELEVATING VOICE DIAGNOSTICS: SAVA UNLEASHES NEW FRONTIERS IN HEALTHY AND PATHOLOGICAL VOICE DETECTION
Author:	ABDUL REHMAN ALTAF, HAIRULNIZAM MAHDIN, AWAIS MAHMOOD, ABDULLAH ALTAF, MUHAMMAD HUSSAIN, SAJID ISLAM
Abstract:	The detection of pathological voices is a pressing and crucial concern that necessitates a thorough exploration of voice signal properties within healthcare settings. While a plethora of voice features exists, typically leveraged through machine learning techniques to distinguish between healthy and pathological voice signals, yet new voice features are required for more promising results. This study introduces a novel voice feature known as the Sum of the Absolute Values of Amplitudes (SAVA) of voice signals. The development of this feature is meticulously detailed in an algorithmic fashion. To ensure its robustness and reliability, a rigorous evaluation technique, known as K-fold cross-validation, has been employed. This approach not only validates the effectiveness of our feature but also provides insights into its stability and generalization capabilities. Utilizing this feature, a novel framework called the SAVA-Based Classifier (SAVABC) has been devised. In addition, the Guassian Naïve Bayes (GuassianNB) machine learning classifier was chosen for implementation. Extensive voice datasets, comprising both healthy and pathological samples from the Saarbrucken Voice Database (SVD), were utilized for experimentation. The results of the simulations are highly promising, achieving an accuracy of 90.21% for vowel classification and 90.65% for sentence classification. Based on these statistics, we assert that the proposed SAVABC system holds the potential for deployment in real-world healthcare settings to detect pathological voices in patients. This system has the capacity to replace manual assessment methods in healthcare, offering substantial convenience to potential patients.
Keywords:	Healthy Voice, Pathological Voice, Machine Learning, Voice Feature, Signal Processing
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	ATTENTION-ENHANCED LSTM MODEL FOR INTRUSION DETECTION IN IMBALANCED NETWORK TRAFFIC DATA
Author:	YAKUB REDDY.K , G.SHANKARLINGAM
Abstract:	Intrusion Detection Systems (IDS) face significant challenges in identifying minority attack classes within imbalanced network traffic, leading to compromised security in critical systems. To address this, we propose an Attention-Enhanced Long Short-Term Memory (AE-LSTM) model that integrates multi-head attention mechanisms with Long Short-Term Memory (LSTM) networks for robust intrusion detection. The model is trained on the NSL-KDD dataset using a comprehensive preprocessing pipeline that includes one-hot encoding, normalization, and SMOTE-based oversampling to mitigate class imbalance, particularly for rare attack types such as User-to-Root (U2R) and Remote-to-Local (R2L). Our architecture incorporates an LSTM layer with multi-head attention, residual connections, and dense layers with dropout regularization. Experimental results demonstrate a classification accuracy of 98.43% and a Top-5 accuracy of 100%. ROC-AUC scores reached 1.00 for most classes, and Precision-Recall analysis confirmed high sensitivity for minority attacks. Visualization via t-SNE revealed distinct inter-class separation. The proposed AE-LSTM model significantly enhances detection performance on imbalanced datasets, presenting a promising approach for next-generation intrusion detection systems (IDS).
Keywords:	Intrusion Detection, LSTM, Attention, SMOT, NSL-KDD
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	CROSS-MODAL ADAPTIVE META-FREE LEARNING FOR SCALABLE CONTINUAL ZERO-SHOT GENERALIZATION
Author:	RAFIAH S B , B PRAJNA
Abstract:	Continual Zero-Shot Learning (CZSL) involves training models to learn sequentially from separate data streams while effectively generalizing to unseen classes without revisiting earlier data. However, conventional approaches often face issues such as catastrophic forgetting and weak generalization, especially under noisy, multimodal, or low-resource conditions. To overcome these limitations, the proposed work introduced Cross-Modal Adaptive Meta-Free Learning (CAMeL) a scalable, task-free learning framework. CAMeL incorporated a Cross-Modal Generative Memory to synthesize both visual and semantic features, ensuring knowledge retention across tasks. It also features a Neural Attribute Synthesizer that generates context-aware prompts, enhancing adaptability to challenging learning conditions. The framework is further optimized through Continual Learning Adaptive Sharpness-Aware Minimization (CLASAM), which flattens the loss landscape to promote stability and generalization. CAMeL effectively supports multimodal learning, reduces forgetting, and handles both zero-shot and few-shot tasks. Experimental results across six benchmarks including CUB, AWA1, and SUN show that CAMeL+CLASAM achieves up to 7.5% higher harmonic mean than existing methods, proving its robustness and scalability.
Keywords:	Continual Learning, Zero-Shot Learning, Sharpness-Aware Optimization, Prompt Learning, Cross-Modal Learning
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	UTILIZING AZURE AUTOMATED MACHINE LEARNING FOR SALES PREDICTION
Author:	HADI SYAHRIAL , FIKRI SALAM , TANTY OKTAVIA
Abstract:	This research assesses the use of Azure Automated Machine Learning (Azure AutoML) at Company XYZ, a prominent flour manufacturer in Indonesia, to overcome the constraints of traditional forecasting techniques. Utilizing the Cross-Industry Standard Process for Data Mining (CRISP-DM) methodology and Azure AutoML's no-code architecture, predictive models were constructed employing five years of historical data. The Voting Ensemble model proved to be optimum, with the Normalized Mean Absolute Error (NMAE) surpassing the Normalized Root Mean Squared Error (NRMSE), attaining an NMAE of 0.21278 in training and enhancing to 0.12118 in testing. The Relative Deviation Averages (RDA) for Products K and Y during one semester were decreased to 10.18% and 0.76%, respectively, surpassing traditional approaches characterized by significant variability. To ascertain dependability, predictions were juxtaposed with actual sales data over a six-month period using semester-based RDA calculations, yielding findings that demonstrated considerable improvement over traditional techniques. Notwithstanding its benefits, Azure AutoML encountered constraints in automated preprocessing activities, necessitating user intervention using Microsoft SQL Server for data cleansing and preparation. The no-code interface allowed non-expert users to deploy models inside Company XYZ's Microsoft environment; nonetheless, successful implementation required a fundamental understanding of statistics and preprocessing methods, including outlier identification, as well as proficiency in MS SQL Server (Transact-SQL). This paper presents a scalable system for improving prediction accuracy via the integration of CRISP-DM methodology, SQL preprocessing, and Azure AutoML, hence facilitating AI deployment in resource-limited settings.
Keywords:	Automated Machine Learning, Azure AutoML, CRISP-DM, Sales Prediction, Time Series
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	INTERNET OF THINGS CONSUMER ELECTRONICS CYBERSECURITY APPROACH BASED ON DEEP LEARNING
Author:	MORARJEE KOLLA, PRAVEENA BAI DESAVATHU, K NARAYANA RAO, NALLA SIVA KUMAR, HANUMANTHA RAO BATTU, JOHN T MESIA DHAS, SUBRAMANYAM KUNISETTI
Abstract:	Protecting linked devices from potential weaknesses and dangers is the primary goal of security in the Internet of Things (IoT) consumer electronics. The collection, transmission, and storage of sensitive information by these smart devices necessitates stringent security measures to prevent hacking, data breaches, and unauthorized access. To ensure the security and privacy of user data, it is essential to utilize strong encryption, secure authentication procedures, and to update software regularly. Users can start and operate drones anytime, anywhere, and they offer a bird's-eye view. Criminals and cybercriminals, however, have begun to exploit drones maliciously. These attacks are extremely dangerous and destructive, and they happen frequently and with a high probability. Therefore, the requirement for investigative work and preventative measures necessitates a desire for protection. Drones equipped with deep learning (DL) for intrusion detection can take control of complex NN structures, enhancing security monitoring in ever-changing outside settings. These drones can fly alone, armed with high-tech sensors and computing power, to scan the sky for dangers like suspicious activity, unapproved people or vehicles, and then label them accordingly. Drones equipped with DL techniques can instantly recognize and respond to complicated patterns and irregularities, paving the way for proactive security measures. An improved method for drone platforms based on mathematical modeling, known as MGOADL-CS, which stands for Mountain Gazelle Optimization with Attention to Deep Learning for Cybersecurity, is presented in this paper. By identifying assaults with the help of optimal DL models, the MGOADL-CS approach seeks to enhance cybersecurity in the drone's environment through the use of BC technology. Starting with input data normalization, the MGOADL-CS method employs a linear scaling normalization (LSN) strategy. When it comes to dimensionality reduction, the MGOADL-CS method relies on an improved tunicate swarm algorithm (ITSA) based feature selection strategy. By the way, cyberattacks are detected and classified using the attention long short-term memory neural network (ALSTM-NN) model. At last, the ALSTM-NN model's hyperparameter values are fine-tuned using the MGO-based hyperparameter tuning procedure. In order to showcase the improved attack detection outcomes of the MGOADL-CS method, a comprehensive simulation set is completed using the NSL dataset. The accuracy rating of 99.71% demonstrated by the performance validation of the MGOADL-CS method was higher than that of previous approaches.
Keywords:	Internet of things, Cons Cybersecurity, Unmanned aerial vehicles, Optimization of mountain gazelles, Deep learning
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	AN INTELLIGENT SYSTEM FOR HAJJ CROWD MANAGEMENT USING DATA MINING TECHNIQUES
Author:	BASHAIR FAHAD , ISLAM R. ABDELMAKSOUD , HAZEM EL-BAKRY
Abstract:	Every year, millions of Muslims come to Mecca to participate in the Hajj pilgrimage. Due to the large number of attendees, there are a variety of logistical and safety concerns that must be resolved. Many times, modern crowd management techniques can be inefficient, resulting in tragic great loss of life events such as the 2015 stampede. In this paper, we explore the use of machine learning methods to predict crowd concentration during the Hajj and Umrah pilgrimages. For this purpose, we apply the Hajj and Umrah Crowd Management dataset available on Kaggle. Our aim is to classify remote sensing crowd density features into three classes, Low, Medium, and High based on certain conditions such as time of the year, weather, and health data. The dataset requires preprocessing such as rescaling, imputation of missing values, and encoding of categorical variables. Feature selection is performed using mutual information to eliminate irrelevant factors that do not aid in predicting crowd density. Hold-out and 5-Fold Cross-Validation techniques are used to train and assess five classification models: Random Forest, K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Decision Trees, and Logistic Regression. The results show that Random Forest performs better than the other models. When feature selection is used, it attains maximum accuracy and F1-scores. The outcomes show how well machine learning predicts crowd density, and Random Forest turns out to be the most dependable model for handling sizable crowds during the Hajj and Umrah.
Keywords:	Hajj Crowd Management, Feature Selection, Data Mining Techniques, Mutual Information, Machine Learning in Crowd Management.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	SELF-ADAPTIVE TRANSIENT GENERALIZED FUNCTIONAL REGRESSIVE CRYPTOSYSTEM BASED AUTHENTICATION FOR SECURED DATA TRANSMISSION IN PERSONAL AREA NETWORK
Author:	R.ABARNA SRI , K.DEVASENAPATHY
Abstract:	Personal Area Network (PAN) is to connect electronic devices. Personal devices namely laptops, wearable devices, and other peripherals employed in PAN.During the communication, PAN is susceptible to different attacks caused by the malicious device within the network. Security plays an important role in preserving the data from different attacks during communication. Secure transmission was discussed with numerous researches. But higher confidentiality was major challenging issues. Self-Adaptive Transient Generalized Functional Regressive Cryptographic Authentication (SATGFRCA) technique is designed to enhance data confidentiality in PANs. The SATGFRCA technique employs the Benaloh Public Key Homomorphic Cryptosystem to achieve improved data confidentiality. In SATGFRCA technique, nodes or devices register their information with the base station (BS) to facilitate authentication. The BS then generates a Self-Adaptive Transient key pair (public and private keys) for each registered node. Upon receiving the key pair, sender node encrypts data packets via public key as well as transmits toward receiver node. Upon receiving the data, the receiver node verifies its authenticity using Generalized Functional Regressive Analysis. Once authenticity is confirmed, the authorized receiver node decrypts the data using its private key. This process ensures data packets remain secure, enhancing both security and confidentiality through the SATGFRCA technique. An experimental evaluation was conducted based on factors such as authentication accuracy, authentication time, data confidentiality rate, and data delivery ratio, considering varying numbers of nodes and data samples. Outcome of SATGFRCA provides higher data confidentiality rate, data delivery ratio, and reduced authentication time. This finding confirms the SATGFRCA technique efficiency and effectiveness compared to existing approaches
Keywords:	Personal Area Network, security, authentication, Benaloh Public Key Homomorphic Cryptosystem, Self-Adaptive Transient key, Generalized Functional regression
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	MESSAGE CONVERSATION BASED SOCIAL ENGINEERING ATTACK DETECTION USING MACHINE LEARNING
Author:	SEAH NI MIN , NOR FAZLIDA MOHD SANI
Abstract:	Social engineering attacks present a major threat in today's interconnected world, exploiting the intricacies of human communication to deceive individuals and extract sensitive information. With the increasing reliance on messaging platforms, communication has become highly informal, often involving colloquial language, abbreviations, and dynamically evolving linguistic styles. These characteristics obscure user intent and make it difficult to identify malicious or deceptive behaviour. Detecting such threats requires a deep understanding of conversational context, which is often lacking in current approaches thereby leaving people vulnerable to subtle social engineering attacks embedded within everyday messages. This study addresses this gap by fine-tuning DistilBERT, a state-of-the-art natural language processing (NLP) model, to detect social engineering attacks in message conversations. Leveraging its ability to understand contextual semantics while maintaining computational efficiency, the model was trained and evaluated using the SMS Spam Collection dataset. The proposed approach achieved a high detection accuracy of 99.46%, outperforming previous models such as SOCIALBERT. While the results demonstrate strong classification performance, limitations include the use of a single text-based dataset and the exclusion of multimodal content such as images and links. Future work should explore more diverse and multilingual datasets, incorporate multimodal detection, and optimise the model further for deployment in resource-constrained environments.
Keywords:	Messages, Social Engineering Attack, NLP, Machine Learning, DistilBERT.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	DEVELOPMENT OF AN EXAM CHEATING DETECTION SYSTEM USING DEEP LEARNING-BASED FACIAL RECOGNITION TECHNOLOGY
Author:	MUHAMMAD SYAHRIANDI ADHANTORO, GANNO TRIBUANA KURNIAJI, RAHAYU FEBRI RIYANTI, HARUN JOKO PRAYITNO, EKO PURNOMO, DIAN ARTHA KUSUMANINGTYAS, ANAM SUTOPO
Abstract:	Cheating in exams, especially in online exams, has become a major challenge for educational institutions. One common form of cheating is the use of exam proxies and disguise attempts using photos or videos. To address this issue, this study develops a fraud detection system based on facial recognition technology using deep learning. The system is designed to automatically identify exam participants, monitor their presence during the exam, and prevent impersonation attempts using images or videos. The research methodology follows the ADDIE model (Analysis, Design, Development, Implementation, Evaluation), encompassing needs analysis, system design, deep learning model development, implementation in real exam scenarios, and system performance evaluation. A facial recognition model based on ResNet-50 is applied to enhance detection accuracy, while a liveness detection feature ensures real-time presence verification of exam participants. The study results indicate that the system achieves an average accuracy of 96.8% in recognizing participants' faces, performing best under normal lighting conditions and frontal face angles. Testing across various scenarios demonstrates the system’s capability to detect fraud, such as the use of exam proxies, impersonation via photos/videos, and participants leaving the exam screen. The implementation of this system has contributed to reducing cheating cases by up to 80% compared to AI-unmonitored exams. The study concludes that facial recognition-based fraud detection systems can enhance transparency, security, and academic fairness in online exams. However, several challenges remain, including technical constraints on participants' devices, potential biases in facial recognition, and ethical and privacy concerns regarding biometric data. This research provides a significant contribution to AI-based academic monitoring and serves as a reference for future developments in technology-driven exam supervision systems.
Keywords:	Deep Learning, Exam Cheating, Face Detection, Face Recognition
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	ENHANCING GROUP DYNAMICS: SOCIAL PRESENCE, COGNITIVE PRESENCE, AND EXPERIENCE IN VOICETHREAD DISCUSSIONS
Author:	SITI NAZLEEN ABDUL RABU, SHARON LEE JIA CHIAN, AHMAD SYAFIQ BIN MOHD NASIR, NURULLIZAM JAMIAT
Abstract:	The study explored undergraduate students’ perceptions of social presence and cognitive presence during collaborative online discussions using VoiceThread. A mixed-methods approach was employed, incorporating an online questionnaire with close- and open-ended questions, analysis of students’ VoiceThread comments, and interviews. Purposive sampling was used to select a cohort of 61 undergraduate education majors. Quantitative data from Likert items were subjected to descriptive and correlation analyses to examine the perceived social presence and cognitive presence, and their interrelationship. Content analysis of commenting modes revealed the frequency of text, audio, video, and doodling comments, supported by open-ended responses to explore students’ preferences. Additionally, thematic analysis was conducted on interviews with a subset of 15 students to examine their experiences using VoiceThread for discussions. Findings indicated that students positively perceived both social and cognitive presences in VoiceThread discussions, with a moderately positive correlation between the two. Text-based comments were most frequently used, followed by audio, while video and doodling were used less often. In terms of its application for group discussion, VoiceThread was valued for its user-friendliness, commenting flexibility, and ability to foster collaborative engagement, though the non-threaded comment structure in the free version was viewed unfavorably. Technical and environmental challenges, such as poor internet connectivity and noisy settings, also hindered effective use. The study suggests the need for clear guidance, practice sessions, and supportive conditions to encourage the use of media-based comments, fostering meaningful interaction and knowledge construction in online group discussions.
Keywords:	Computer-mediated Communication (CMC), VoiceThread, Online Discussion, Community of Inquiry, Social Presence, Cognitive Presence, Commenting Modes, Experience
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	GHOSTFREAK: ENHANCED STEGANOGRAPHY IN PRINT-SCAN ENVIRONMENTS USING DEEP LEARNING
Author:	S. KOMAL KOUR, SOHAN GUNDOJU, SREE VAIBHAV DUVVURI, DR. T ADILAKSHMI, T JALAJA
Abstract:	GhostFreak, a novel deep steganography framework designed to address the challenges inherent in print-scan pipelines. Traditional steganographic approaches often struggle with the distortions introduced during the printing and scanning process, leading to significant degradation in the quality and recoverability of embedded data. To overcome these limitations, GhostFreak strategically leverages three distinct color spaces—RGB for digital displays, HSI for human visual perception, and CMYK for printing—to achieve robust and imperceptible data embedding while maintaining resilience against real-world distortions. Framework extends prior research on print-scan steganography by integrating a U-Net GAN architecture, which processes a concatenated tensor of multi-color representations alongside a secret code to generate a residual image. This approach ensures that the hidden information remains intact while preserving the quality of the Stego image. A key innovation of GhostFreak is its dynamically weighted loss function, which balances multiple loss components—secret loss, secret decay loss, LPIPS (perceptual similarity) loss, residual loss, and edge loss—to optimize the model's performance across different training phases. During initial training stages, the model prioritizes accurate encoding of the secret message, whereas in later stages, the emphasis shifts towards ensuring high-fidelity image reconstruction. We validate the effectiveness of Ghost Freak through comprehensive experiments on a large-scale image dataset, demonstrating that our method withstands the degradations introduced by print-scan operations. Comparative evaluations against prior steganographic approaches highlight significant improvements in terms of imperceptibility, robustness, and recoverability of the embedded data. The results indicate that GhostFreak is a promising advancement in deep steganography, offering a practical and resilient solution for secure data embedding in printed media.
Keywords:	Image Steganography, Deep Learning, Print-Scan, Color Spaces, U-Net GAN, Dynamic Loss Weighting, Error Correction.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	EXPLORING QUANTUM MACHINE LEARNING ALGORITHMS FOR ENHANCED CHEMICAL TOXICITY CLASSIFICATION
Author:	KALIDINDI VENKATESWARA RAO, KUNJAMNAGESWARA RAO, GOKURUBOYINA SITARATNAM
Abstract:	Classical computational approaches are steadily improving in their ability to predict molecular properties based on their sequences or structure, which is essential in drug discovery, irrespective of its many challenges. The major challenge is to use computational approaches to identify the properties, such as toxicity, solubility, etc. (ADMET), of the drug targets without costing much time. Machine learning and deep learning methods are employed for predicting ADMET properties. The proposed approach combines quantum computing with machine learning for the binary task of toxicity prediction based on the SMILES data. This Quantum Machine Learning (QML) approach involves a novel five-step procedure that converts SMILES to their respective molecular fingerprints, which are in turn dimensionally reduced, consequently forming four qubits. These qubits are inputs to the QML models, such as the Quantum SVC (QSVC), Variational Quantum Circuits (VQC), and Quantum AutoEncoders (QAE). The models are trained and optimized by using different optimizers and then evaluated based on the accuracy metric. The QAE model outperformed the remaining model by achieving an accuracy of 99%.
Keywords:	ADMET, SMILES, QML, QAE, QSVC, VQC
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	IOT SYSTEM USING DEEP LEARNING FOR REAL-TIME HEALTH MONITORING AND ALSO PRIMARY RECOGNITION OF HEALTH PROBLEMS
Author:	KASI VENKATA KIRAN , T SRINIVASA RAO
Abstract:	The significance of remote health monitoring is very important to enhancing patient care and decreasing healthcare expenses has grown in current years owed towards the rising occurrence of chronic illnesses in addition the ageing population. There has been a lot of buzz lately about Internet of Things (IoT) as a possible solution for remote health observation and monitoring. Systems built on the Internet of Things may gather and process a plethora of physiological data, by means of heart rates, temperature readings, blood oxygen stages, then electrocardiogram (ECG) signals, and then provide doctors immediate input on how to proceed. In home healthcare settings, this study suggests the Internet of Things (IoT) founded scheme for remotely detecting, monitoring then primary identification of health concerns. The system is made up of three different kinds of sensors: a MAX30100 that measures heart rate and blood oxygen levels, an AD8232 that records ECG signals, and an MLX90614 that takes temperature without touching the skin. We use the MQTT protocol towards direct the data we've gathered to a main server. Then the main server habits the deep learning architecture that has already been trained using an attention layer in a convolutional neural network to categorise possible illnesses. In addition to normal heart rate, the system can distinguish between five distinct types of heartbeats based on electrocardiogram (ECG) data: premature ventricular contraction, supraventricular premature beatings, unclassifiable beatings, and synthesis of ventricular. Not only that, the device tells you whether the patient's oxygen levels and heart rate are normal or not. If serious irregularities are found, the system will link the user to the closest doctor for additional diagnostics.
Keywords:	Deep Learning, Internet of Things (IoT), Convolutional Neural network, Sensor, Health Monitoring
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	MULTI-USER EMOTION RECOGNITION IN CROWDED SCENES VIA ENHANCED PARTICLE SWARM OPTIMIZED RECURRENT NEURAL NETWORKS
Author:	BHAGYASHREE DHARASKAR, PHANI KUMAR KANURI, M. SHARWANI, SANDEEP RASKAR, DR. N. SREE DIVYA, DR. KANAKA DURGA HANUMANTHU, DR. MANISHA P. MALI, RAMESH BABU PITTALA, M. NAGABHUSHANA RAO
Abstract:	Psychological group level emotion recognition (GER) is significant because it facilitates understanding and identifying the behavior of people in large congregations, organizations, and other facilities that require surveillance. Some issues include occlusions, dynamic facial expressions, variation in pose, and even low-resolution faces. A new framework of combining the EPSO (Enhanced Particle Swarm Optimization) algorithm for identifying significant features to tackle these challenges and using RNN to learn the sequences involved. The two-fold process involves feature reduction followed by meaningful group emotion classification, taking into consideration the temporal dynamics of emotions. The Acted Facial Expressions in the Wild (AFEW) dataset is used to validate the proposed EPSO-RNN model by comparing it with the baseline methods, such as CNN, SVM, and VGG-16. The experimental findings reveal better EPSO-RNN results in different measures of recognizing group-level emotions. In order to enhance the capabilities of the existing PSO, the proposed EPSO-RNN framework finally combines the enhancement of the feature space optimization and sequential emotion modeling by using recurrent neural network structures. Unlike existing methods that primarily rely on deep convolutional architectures or conventional classifiers with limited adaptability to real-world group settings, this study fills a significant gap by introducing a hybrid EPSO-RNN model that jointly optimizes feature selection and temporal emotion learning. The novelty lies in leveraging enhanced particle swarm optimization to distill high-impact features from noisy group environments, followed by RNN-driven modeling of emotional evolution over time. Experimental validation on the AFEW dataset shows a substantial improvement in classification accuracy and F1-score, outperforming traditional CNN, SVM, and VGG-16 baselines. These findings confirm the framework's potential for robust and scalable deployment in practical surveillance and organizational emotion analytics scenarios. The abstract clearly states that the study introduces a hybrid EPSO-RNN model, offering new insight into combining feature selection and temporal learning for improved group emotion recognition in real-world conditions.
Keywords:	Group-level Emotion Recognition, Recurrent Neural Network (RNN), Enhanced Particle Swarm Optimization (EPSO), Feature Selection, Deep Learning, Crowd Emotion Analysis, Video Frame Processing,
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	DEEP LEARNING-DRIVEN AUTOMATED IOT DEVICE IDENTIFICATION USING FULL PACKET DATA
Author:	SINGAMANENI KRISHNAPRIYA
Abstract:	The rapid proliferation of Internet of Things (IoT) devices has introduced significant challenges in network security, device management, and traffic monitoring. Accurate and automated IoT device iden- tification is critical for ensuring secure communication, anomaly detection, and enforcing access control policies. Traditional identification methods, relying on static rule-based approaches or shallow learning techniques, struggle with the increasing diversity and evolving communication patterns of IoT devices. In this study, we propose a novel deep learning-driven framework that leverages full packet data analysis to achieve robust and scalable IoT device identification. The framework integrates Convolutional Neu- ral Networks (CNNs) for spatial feature extraction and Long Short-Term Memory (LSTM) networks for temporal pattern learning, enabling it to effectively capture packet header structures, payload distributions, and sequential dependencies in IoT traffic. Additionally, dropout regularization is employed to enhance generalization and mitigate overfitting, ensuring resilience across heterogeneous IoT environments. The proposed method is evaluated using benchmark IoT datasets, including UNSW IoT-23 and NB-IoT, which demonstrate superior classification accuracy, scalability, and adaptability compared to existing approaches. Experimental results highlight the effectiveness of hybrid deep learning models in IoT security, achieving high precision and low false positive rates in device identification. This research underscores the potential of full packet data-driven deep learning approaches to fortify IoT network defenses and advance next- generation automated cybersecurity solutions.
Keywords:	Deep Learning,Full Packet Data Analysis,Convolutional Neural Networks (CNN),Long Short Term Memory (LSTM),Anomaly Detection
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	AI-POWERED JOB APPLICATION MANAGEMENT FOR APPLICANTS
Author:	ZI QING CHEW , NOR FAZLIDA MOHD SANI
Abstract:	During the job search process, a resume serves as a job applicant's first impression and will determine whether they will progress in the hiring process. One of the major challenges job seekers faced is ensuring that their resumes align with job requirements, as qualification mismatches can lead to missed job opportunities. The advancement of Applicant Tracking Systems (ATS) presents an additional challenge, as resumes lacking relevant keywords or proper formatting may be automatically rejected by ATS during the automated resume screening phase. To address these challenges, an intelligent job application management system has been developed, contributing to IT research through novel integration of transformer-based NER models with interpretable job-resume matching metrics for analyzing and improving resume contents. The methodology involved a semi-automated annotation process to prepare the annotated resume dataset, followed by training Named Entity Recognition (NER) models using the spaCy and Flair libraries. These trained models were evaluated and compared based on precision, recall, and F1-score metrics. The job-resume matching score was calculated by comparing TF-IDF vectorization of NER-extracted skills from resumes and job descriptions using cosine similarity, followed by normalization with the Sigmoid function. Experimental results showed that the spaCy model achieved an F1-score of 85.71%, outperforming the Flair NER model, which achieved an F1-score of 79.92%. This research advances IT applications in human resource technology by assisting job applicants in enhancing and tailoring their resumes to better match desired job roles, increasing their chances of passing ATS resume scans and being shortlisted for job interviews.
Keywords:	Resume; Job-Resume Matching; NER; Cosine Similarity; Natural Language Processing.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	ADVANCED LUNG CANCER DETECTION MODEL WITH U-NET-BASED SEGMENTATION, ARI-TFMOA FEATURE OPTIMIZATION AND MACHINE LEARNING CLASSIFICATION
Author:	SATISHKUMAR PATNALA, ANNEMNEEDI LAKSHMANARAO , SUJAN BABU VADDE , KESHETTI SREEKALA , MOUPALI ROY, Dr. ASMITA MANNA
Abstract:	Lung cancer is among the deadliest diseases globally, highlighting the importance of early and accurate detection to enhance survival chances. Traditional deep learning models face challenges such as high computational costs and lack of interpretability, while conventional machine learning classifiers require high-quality feature representations for optimal performance. To address these challenges, this study proposes an advanced lung cancer detection model integrating segmentation, feature optimization and classification using both machine learning and deep learning models. The framework employs AMTUnet++-ASPP for abnormality segmentation, ensuring accurate lung nodule detection from CT scan images. The segmented tumor regions undergo feature optimization using Adaptive Radiant Inertia Tuned Fuzzy Multi-Objective Algorithm (ARI-TFMOA), which refines extracted features and eliminates redundant information. The optimized feature vectors are then classified using multiple models, including ML algorithms and an artificial neural network (ANN) to distinguish between normal, benign, and malignant lung conditions. To evaluate the effectiveness of the proposed model, a comparative analysis is conducted by implementing three different approaches: a baseline CNN for direct classification, a U-Net-based segmentation followed by ML classification, and the proposed AMTUnet++-ASPP segmentation with ARI-TFMOA optimized classification using both ML models and ANN. The results demonstrate that the integration of advanced segmentation and optimized feature selection significantly enhances classification performance. In particular, the ANN model outperformed other classifiers, leading to superior accuracy, precision, and recall values. This hybrid approach, combining deep learning-based segmentation, feature optimization, and deep learning classification, provides a computationally efficient and clinically interpretable solution for AI-assisted lung cancer diagnosis, reinforcing its potential for early detection and decision support for radiologists.
Keywords:	Lung Cancer Detection, CT scan images, AMTUnet++, ARI-TFMOA, Machine Learning, Deep Learning.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	LATENCY-AWARE NETWORK SLICING USING DISTRIBUTED RESOURCE ALLOCATION AND REGIONAL ORCHESTRATION
Author:	SAIF SAAD ALNUAIMI
Abstract:	Network slicing has emerged as a fundamental enabler for delivering diverse services with heterogeneous requirements in next-generation communication systems. However, most existing approaches focus on either bandwidth or computational resource allocation in isolation, often relying on centralized architectures that struggle with latency, scalability, and information privacy across distributed network components. This paper addresses this gap by proposing a novel distributed network slicing architecture that jointly optimizes bandwidth and compute resources. The core innovation is the introduction of a regional orchestrator (RO) a new control plane entity positioned between base stations (BSs) and cloud nodes to coordinate localized resource allocation while preserving system privacy and scalability. We develop a distributed resource allocation algorithm based on the splitting model (dra-SM) to efficiently manage joint resource distribution without centralized control. Simulation results show that our approach significantly reduces overall network latency by approximately 15%—compared to single-resource slicing, while also achieving faster convergence and service-specific latency guarantees. This work contributes a scalable, low-latency solution for real-world deployment of joint network slicing across decentralized infrastructures.
Keywords:	Data Slicing, Cloud, Network Slicing, Bandwidth Consideration, Resource Allocation.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	AI AND IOT-DRIVEN SMART CITIES: REVOLUTIONIZING ENERGY EFFICIENCY AND OPTIMIZING TRAFFIC FLOW FOR SUSTAINABLE URBAN LIVING
Author:	CH GANGADHAR, FRANCIS MULAGANI, KOLLI SRINU, KOLLURU SURESH BABU, ANIL KUMAR KATRAGADDA, K SWATHI, T MURALIDHARA RAO, CH CHANDRA MOHAN
Abstract:	This paper investigates the potential of using AI and IoT to improve energy efficiency and traffic management in smart cities. This paper proposes an innovative framework, AI-IoT, that combines real-time traffic control with a real-time energy management system to alleviate urban congestion, mitigate energy loss, and reduce carbon emissions. The models can regulate signaling around street junctions and energy distribution in real-time by applying sensors, IoT devices for data collection, and AI algorithms for immediate decisions. The model was validated with simulations and showed a significant reduction in traffic impact on urban sustainability (12% less energy consumption, 20% less traffic congestion, and 17% less CO2 emissions). The findings show how AI and IoT can be harnessed to build smarter and greener cities, delivering a repeatable approach to designing smarter cities in the future.
Keywords:	Smart Cities, Artificial Intelligence, Internet of Things, Energy Efficiency, Traffic Flow
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	PHOTOMETRIC WEIGHT FORMULATED TRILATERAL FILTERS FOR ENHANCED PERCEPTUAL QUALITY IN BIOMEDICAL IMAGING
Author:	SUDAGANI JYOTHI , P. MUTHU KRISHNAMMAL
Abstract:	This paper presents a novel computational framework for biomedical image enhancement through the development of Photometric Weight Formulated Trilateral Filters (PWFTF), advancing adaptive image processing algorithms for clinical applications. Unlike conventional bilateral and standard trilateral filters, our proposed method incorporates adaptive photometric weights that respond dynamically to local image characteristics across different imaging modalities. We evaluate the performance of our method on Ultrasound, X-Ray, and MRI datasets, demonstrating significant improvements in noise reduction, edge preservation, and overall perceptual quality. Comprehensive objective assessments using established Image Quality Assessment (IQA) metrics show that our PWFTF method outperforms state-of-the-art filtering techniques by an average of 17.3% in PSNR, 12.6% in SSIM, and 9.8% in FSIM across all tested modalities. This framework advances IT research by enabling scalable, computationally efficient image processing for real-time clinical diagnostics, reducing overhead compared to learning-based methods. The proposed filter demonstrates particular effectiveness in preserving diagnostically significant features while suppressing noise in low-contrast regions, making it suitable for clinical applications requiring high diagnostic accuracy.
Keywords:	Biomedical Image Processing, Trilateral Filtering, Ultrasound Imaging, X-Ray Imaging, MRI.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text

Title:	ENHANCED WIDE-AND-DEEP NEURAL NETWORKS FOR ROBUST MALWARE DETECTION USING OPCODE FEATURES AND ZERO-DAY ANALYSIS
Author:	G BALA KRISHNA, K.C. SREEDHAR, V. PRADEEP KUMAR, SAIROHITH,THUMMARAKOTI, D. SANDHYA RANI, DR. MUMMADI RAMACHANDRA
Abstract:	Detecting malware is a vital cybersecurity challenge, necessitating more enhanced means of detecting and fighting such threats. To classify malware families and detect previously unseen malware, the opcode sequences and metadata are used with proposed architecture of an Enhanced Wide-and-Deep Neural Network (EWDNN). Combining wide components for explicit feature interactions and deep components for abstract feature representations to structure the EWDNN, we propose a model that outperforms all baselines with an accuracy of 65.46%, precision of 63.96%, recall of 65.46% and an F1-score of 64.15%. The proposed methodology exhibits scalability, robustness, and generalization, and outperforms other traditional approaches, including Support Vector Machines, K-Nearest Neighbors, and standalone Deep Neural Networks. Extensive experiments demonstrate that the model is capable of real world malware detection, especially for zero day threats, showing its usefulness in real world dynamic cybersecurity situations.
Keywords:	Malware Detection, Enhanced Wide-and-Deep Neural Network (EWDNN), Opcode Sequences, Zero-Day Malware Detection, Scalability, Robustness, Machine Learning, Cybersecurity.
Source:	Journal of Theoretical and Applied Information Technology 31^st July 2025 -- Vol. 103. No. 14-- 2025
Full Text