		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 /submit_paper.php in an MSWord, Pdf or compatible format so that they may be evaluated for publication in the upcoming issue. This journal uses a blinded review process; please remember to include all your personal identifiable information in the manuscript before submitting it for review, we will edit the necessary information at our side. Submissions to JATIT should be full research / review papers (properly indicated below main title).

Journal of Theoretical and Applied Information Technology
March 2025 | Vol. 103 No.6

Title:	DEEP LEARNING-DRIVEN FORECASTING MODELS FOR IOT DATA IN CLOUD COMPUTING ENVIRONMENTS: LEVERAGING TEMPORAL CONVOLUTIONAL NETWORKS
Author:	OTHMAN EMRAN ABOULQASSIM, FARHAT EMBARAK, JAYASHREE S, ABDULHAMID ELTHEEB
Abstract:	Data-driven techniques for machines tool wear detection and forecasting have gained prominence in the past several years. The study investigates how well Temporal Convolutional Networks perform in cloud computing contexts for IoT data prediction. Because TCNs are good at capturing temporal patterns and long-term relationships, they are useful for time-series forecasting problems. Utilizing convolutional layers, TCNs differ from conventional Recurrent Neural Networks in that they analyse data in addition, to enhancing adaptability and decreasing training time. Dilated convolutions are included in TCNs to further improve their capacity to identify trends over long periods without adding to the computational complexity, which makes them appropriate for connections that last and recurrent trends in IoT data. The study shows that TCNs perform better than existing models like RNN, LSTM, GRU-LSTM, and CNN-LSTM in terms of metrics like R^2 alongside and Mean Absolute Percentage Error. The study was conducted on a Python platform running Windows 11. TCNs attained an MAE of 98.7%, RMSE of 97.6%, MAPE of 98.0%, and R^2 of 97.7%, according to the results. Although the error metrics are greater, the significant R^2 value suggests a strong model fit. The study draws attention to many problems with TCNs, such as the requirement for large labelled datasets, understanding, data quality, and computationally demanding requirements. The study also highlights how scalability and flexibility offered by cloud platforms enable effective management of massive IoT data streams and real-time analysis. The results indicate that TCNs may greatly increase resource use and forecasting accuracy in IoT-cloud environmental systems, but more development and study are required to fully realize their capabilities.
Keywords:	IOT Data Forecasting, Cloud Computing In Iot, Temporal Convolutional Network (Tcns), Deep Learning, Dilated Convolutions, Machine Tool Wear Detection
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	IMPROVED DEEP LEARNING WITH SELF-ADAPTIVE ALGORITHMS FOR ACCURATE STRESS DETECTION: CASCADED CNN_BILSTM_GRU METHOD
Author:	FATEH BAHADUR KUNWAR, RAKESH KUMAR YADAV, HITENDRA SINGH
Abstract:	Stress detection is crucial in various fields due to its significant negative effects on individuals and groups. Eradicating stress is difficult, thus the need to manage its physical and mental consequences. Current stress detection methods are ineffective and require enhancement. Traditional approaches struggle to accurately detect stress, particularly with complex and diverse data. The paper introduces a new model, Cascaded NN_BiLSTM_GRU with self-adaptive Walrus Optimization Algorithm (SA-WaOA), to address stress detection inefficiency by using adaptive techniques. The aim is to develop a dependable deep-learning model for early identification and support in mental stress intervention and resource allocation to enhance individual well-being. This newly proposed structure involves a cascaded CNN_BiLSTM_GRU where the dilation rate is adaptive by the Weibull distribution function, adaptive dropout determined by Cumulative Distribution Function (CDF), and establishment of adaptive loss function by exploiting the Bernoulli distribution. The Gated Recurrent Unit (GRU), in a cascaded CNN_BiLSTM_GRU, has been utilized for learning the spectral and temporal features. The model using Python platform shows improved stress detection accuracy. The research contributes significantly to the field of IT by introducing an innovative deep-learning model that leverages adaptive mechanisms for enhanced stress detection, setting a new benchmark in mental health monitoring and intervention. The proposed approach offers a robust framework for handling complex and diverse stress-related data, thereby improving the accuracy and efficiency of stress detection systems.Model's performance evaluated using various metrics. Results show high accuracy in different datasets and learning rates. EEG Feature dataset: 95.05% accuracy (70/30), 96.33% accuracy (80/20). Emotion dataset: 95.51% accuracy (70/30), 96.28% accuracy (80/20). Stress Detection dataset: 95.95% accuracy (70/30), 96.65% accuracy (80/20). DASPS dataset: 96.71% accuracy (70/30), 97.71% accuracy (80/20).
Keywords:	Cascaded CNN_Bilstm_GRU; Self-Adaptive Walrus Optimization Algorithm; Weibull Distribution Function; Cumulative Distribution Function; Bernoulli Distribution Function.
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	ROLES AND CHALLENGES OF BLOCKCHAIN TECHNOLOGY ADOPTION IN ACCOUNTING AND AUDITING
Author:	MARCELLINE AUDI LEO , MARSHA MARGARETHA , LUSIANAH
Abstract:	The transformation of digital technology has had a significant impact on various industries, one of which is in accounting and auditing practices, where blockchain technology offers a potential in increasing transparency, security, process efficiency, and data reliability in accounting and auditing. This study examines various literatures to evaluate the benefits and constraints faced in implementing blockchain technology adoption in accounting and auditing through a literature review. Blockchain, as a decentralized technology that offers high transparency and security, has the potential to develop accounting and auditing practices by improving efficiency, effectiveness, security, accuracy, and reliability of data. However, there are several significant challenges that still hinder the widespread adoption of this technology, such as regulatory constraints and cyber and hacker threats. This study contributes to providing a comprehensive understanding of how blockchain technology can be integrated into accounting and auditing processes. The findings of this study are expected to help practitioners and researchers in developing more detailed and in-depth effective strategies by utilizing the potential of blockchain technology to increase transparency, efficiency, security, and trust in financial information management.
Keywords:	Blockchain, Accounting, Auditing, Technology, Literature
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	ACCESSIBILITY GAPS IN OMANI HOSPITAL WEBSITES: A WCAG 2.1 COMPLIANCE STUDY
Author:	MOHAMMAD ABU KAUSAR , MOHAMMAD NASAR , SALLAM OSMAN FAGEERI
Abstract:	In the digital age, ensuring web accessibility in the healthcare sector is critical for inclusivity, especially for individuals with disabilities. This study evaluates the accessibility compliance of top-ranked hospital websites in Oman against the Web Content Accessibility Guidelines (WCAG) 2.1 standards. Through the use of automated tools (e.g. WAVE, TAW and EIII) accessibility metrics were analyzed to find common accessibility problems such as: missing alternative text, low contrast ratios between background and foreground, and operability issues. Public and private sector websites of the top six hospitals formed the sample. Results show significant differences, with private hospitals being more prone to accessibility violations than public hospitals. Common categories of accessibility mistakes included information and relationships, non-text content, labels or instructions, headings and labels and keyboard accessibility. The study concludes that most hospital websites in Oman generally fail to meet web accessibility standards, restricting the functionality available to many users. This study highlights the need for an organized web accessibility criteria framework for the healthcare sector in Oman and contributes valuable perspectives toward a more digitally inclusive future as a part of the Vision 2040 goal.
Keywords:	Web Accessibility, WCAG 2.1, Healthcare Websites, Digital Inclusivity, Accessibility Compliance, Vision 2040
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	USING SENTENCE TRANSFORMERS FOR SELF-ASSESSMENT IN DIGITAL TRANSFORMATION
Author:	Bachira Abou El Karam , Tarik Fissaa , Rabia Marghoubi
Abstract:	Digital transformation requires rethinking a company’s organization to identify the necessary changes for implementing digital initiatives. It goes beyond technology, encompassing corporate strategy and impacting organizational culture, employee involvement, customer orientation, and business models. To embark on a digital transformation project, companies must first assess their current state regarding strategy, digital maturity, and organizational culture. Existing evaluation methods rely either on consulting services, which are effective but costly for SMEs, or closed-response questionnaires, which are quicker and standardized but limit expression, potentially introducing biases. To address these challenges and find a compromise between the affordability and the precision, this paper proposes an automated self-assessment approach based on open-ended responses, leveraging Sentence Transformers to evaluate and score SMEs' current state. Since the aforementioned evaluation requires high precision due to strategic decisions and investments that are resulting, and characterized by the diversity of unrestricted responses, particularly in a francophone context where cultural and linguistic nuances can significantly influence results, the approach must be tested and compared to the manual method which is often considered as a reference. To achieve this, a case study was conducted on a Moroccan SME that had previously been audited manually by a consulting firm, and the open-ended responses from this audit were subsequently analyzed automatically using a Sentence Transformer-based method as well as using statistical techniques: TF-IDF and LSA. The results demonstrated a strong alignment between the proposed approach and expert evaluations, confirming its effectiveness as a cost-efficient and scalable solution for SMEs, while outperforming other evaluated methods.
Keywords:	Digital Transformation, Self-Assessment, Neural network transformers, NLP, Semantic Similarity, Decision Making.
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	ENHANCING POWER SYSTEM SECURITY WITH A HYBRID SATS ALGORITHM FOR OPTIMAL POWER FLOW
Author:	KUMAR CHERUKUPALLI , BADDU NAIK BHUKYA , PADMANABHA RAJU CHINDA
Abstract:	The security and performance of electricity systems are the primary concerns in their planning and operation. It is imperative to establish appropriate procedures for the maintenance and enhancement of security within the power system. This research introduces a hybrid simulated annealing and tabu search (Hybrid SATS) approach to address the security-constrained optimal power flow problem. The main aim of the research is to improve power system security and reduce generator fuel expenses. Contingency ranking is employed to determine line outages. The Hybrid SATS approach effectively alleviates line flow limit breaches during various single line outages, maintaining power flows within secure parameters. To evaluate the efficacy of the proposed Hybrid SATS approach, simulation experiments are conducted on the standard IEEE 30 bus, and the results are compared with those of the SA and TS methods.
Keywords:	Security Constrained Optimal Power Flow, Simulated Annealing, Tabu Search, Contingency.
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	A STRUCTURED TRACEABILITY APPROACH FOR TRANSFORMING REQUIREMENTS INTO CLASS DIAGRAMS
Author:	KELETSO J. LETSHOLO
Abstract:	In software engineering, requirement traceability is a critical factor in ensuring that software systems comply with user requirements, enabling effective management and verification throughout the development lifecycle. However, due to inherent ambiguities and inconsistencies in user-specified requirements, typically expressed in natural language, establishing accurate traceability remains a significant challenge. This study introduces a novel, automated approach to requirement traceability, addressing this challenge by transforming natural language requirement specifications (NLRs) into class diagrams while simultaneously generating traceability links. The proposed approach leverages Natural Language Processing (NLP) techniques and Semantic Object Models (SOMs), a structured and reusable pattern-based method that enhances the accuracy and consistency of traceability links. To validate this approach, the TRAM application was developed and evaluated against manually produced class diagrams from requirements engineering experts. Precision and recall metrics were used to assess the accuracy and completeness of the generated traceability links. Results indicate that TRAM achieves high recall (0.60–0.96), demonstrating its effectiveness in capturing relevant elements, although precision (0.25–0.51) remains a challenge due to the integration of predefined elements through SOM patterns. These findings highlight the contribution of this research in advancing automated traceability solutions, reducing manual effort, and improving consistency in requirements engineering. Additionally, the structured use of SOMs suggests that this methodology can be extended beyond class diagrams to other software artifacts, broadening its applicability in software engineering.
Keywords:	Class Diagram; Natural Language Processing; Requirements Engineering; Requirements Traceability; Semantic Object Model.
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	IMPLEMENTING NEUROMORPHIC COMPUTING USING NEURAL NETWORK TECHNOLOGY FOR DYNAMIC OBJECT DETECTION AND OCCLUSION HANDLING IN AR IMAGES
Author:	ARUNA THETHALI, MANDAVA KRANTHI KIRAN
Abstract:	In recent times, people, objects, and other real-time elements have played a major role in real-time videos. Most scenes are taken behind natural objects like trees, grid wires, wireframes, etc. Identifying, detecting, and tracking the objects from real-time videos is inaccurate since the objects are over-placed on one another. Some earlier research has focused on providing occlusion detection in face recognition to identify human faces and obtained only 80% accuracy. In real-time industries like virtual and augmented reality, video occlusion is high, and objects must be identified accurately. This paper focused on occlusion removal and object detection to improve visualization accuracy. To increase object detection accuracy without occlusion, this paper proposes an SNN-based neuromorphic system to detect objects from occluded images. The mask-RCNN model is applied to segment the input samples before detection to detect the objects accurately. The overall workflow of the proposed SNN-based model uses four steps: data pre-processing, feature extraction, data segmentation, and detection. The entire model is implemented and experimented with AR video images, and the results are verified. The output shows that SNN and MRCNN increased the overall efficiency in object detection and visualization, occlusion removal, reduced time, and computational complexities.
Keywords:	Neuromorphic Computing, Spiking Neural Networks, MR-CNN, Occlusion Removal, Object Detection.
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	OPTIMIZING UNDERWATER IMAGE QUALITY: ADVANCED TECHNIQUES FOR HAZE REMOVAL, CONTRAST ENHANCEMENT, AND COLOR CORRECTION
Author:	PRANJIT LAHON , DR. RANJAN SARMAH
Abstract:	The enhancement of underwater imagery is essential due to the numerous challenges such as turbidity and light scattering that cause poor visibility, reduced contrast, and color distortion. This study introduces a methodology aimed at improving underwater images through techniques like contrast enhancement, haze removal, and color correction. Core methods employed include median filtering for noise reduction, gamma correction for brightness adjustment, the Dark Channel Prior (DCP) for haze removal, and Contrast Limited Adaptive Histogram Equalization (CLAHE) for contrast enhancement. We conducted experimental evaluations on the "SHARK" test image using several algorithms, including Histogram Equalization (HE), Spectral Information Divergence (SID), Fusion (FU), and Image Analysis (IA). These algorithms were assessed based on the Underwater Image Quality Measure (UIQM) and Processing Time (PT). The proposed method achieved a leading UIQM score of 1.751 with the shortest processing time of 0.452 seconds, surpassing other techniques in both image quality and efficiency. This high-quality enhancement coupled with fast processing renders the proposed method particularly suitable for real-time and resource-limited scenarios. The proposed methodology notably enhances the clarity, contrast, and color accuracy of underwater images, making them more effective for applications in marine research and underwater exploration. The significant improvements in underwater image enhancement techniques demonstrated in this study highlights the potential for further advancements in various fields, addressing the unique challenges posed by underwater imaging.
Keywords:	Contrast enhancement, Haze removal, Color correction, Dark Channel Prior (DCP), Contrast Limited Adaptive Histogram Equalization (CLAHE), Underwater Image Quality Measure (UIQM), Processing time (PT).
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	GEOPOLITICAL CONSEQUENCES OF ARTIFICIAL INTELLIGENCE GOVERNANCE
Author:	MARINA SHULGA , BOHDAN KACHMAR , ANDREY SHARYPIN ,SERHII STAVROYANY, ANDRII TIMCHENKO
Abstract:	The article critically examines the geopolitical implications of artificial intelligence (AI) governance, highlighting its role in reshaping international power dynamics. This study addresses the gap in the literature regarding AI’s impact on national security, economic dominance, and political control. By analyzing policy frameworks and global AI leadership strategies, the study provides a novel perspective on AI governance as a geopolitical tool. Findings indicate that AI enhances technological sovereignty, strengthens defense capabilities, and contributes to cyber and economic conflicts. The conclusions emphasize the necessity of a comprehensive global regulatory framework to mitigate AI-related risks and foster international cooperation. This study contributes new knowledge by demonstrating how AI governance influences strategic geopolitical stability and security.
Keywords:	AI Governance, Geopolitical Impact, AI Regulation, National Security, Economic Power, AI Ethics
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	SOFTWARE PROJECT MONITORING USING MILESTONES AND PERT-AN EMPIRICAL ANALYSIS
Author:	DR. GADALA SUHASINI, DR S. SAI KUMAR,3MONICA KALBANDE, DR. P. PUNITHA, DR HARI KISHAN CHAPALA, CHINNAM SABITHA, CHETLA CHANDRA MOHAN
Abstract:	This paper concentrates on monitoring projects through the PERT and MILESTONES. Complex, multilayered and distributed projects require a series of activities, some of which must be preferred sequentially and others parallely. This collection of series and parallel tasks can be modeled as a network. PERT is statistical technique applied to such a networks. In this paper we attempted to simulate the PERT networks and graphically represented the tasks along with them inter dependencies. Here project identifies the critical and non-critical tasks and evaluates the critical path to determine which tasks have an impact on the schedule. Once a project has advanced to the phase of performance, the focus shifts from the discovery to tracking and reviewing it. MILESTONES are used to track the progress of the project at different stages and the PERT chart, on continual basis. This paper integrated both these techniques for an efficient and easier monitoring. In order to reap the results of the project sooner, we gave a provision to reduce the scheduled completion time with minimum cost burden. This can be achieved by assigning more labor and resources to the various activities.
Keywords:	PERT, MILESTONES, Project, Resources, Schedule.
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	BRIDGING THE IT SKILL GAP WITH INDUSTRY DEMANDS: AN AI-DRIVEN TEXT MINING APPROACH TO JOB MARKET TRENDS USING LARGE LANGUAGE MODEL
Author:	TODSANAI CHUMWATANA , AYE KHIN KHIN HPONE
Abstract:	The growing influence of technological advancements and the complexity of business requirements have revolutionized the IT industry underscoring the critical need for a comprehensive understanding of the skills and tools most valued and prioritized by employers. However, identifying these in-demand competencies remains a challenge given the dynamic nature of the job market. With previous studies exploring IT skill demands and very few studies utilizing AI-driven text-mining techniques to systematically analyze a large number of job advertisements, this research aims to address the ongoing gap between individual competencies and industry requirements. Leveraging text mining techniques, natural language processing (NLP), and large language models (LLMs) such as GPT-based AI model to enhance keyword extraction, this study provides the workflow process for generating knowledge-based insights using Term Frequency-Inverse Document Frequency (TF-IDF) scoring from a collection of IT job advertisements. The study identified vocabularies related to key skills and tools, by effectively summarizing the occurrences in job postings and revealing key terms that characterize specific roles and industry demands across IT sectors such as Software Development, DevOps, Cloud, Database Administration, Business Analysis, and Business Intelligence. Key findings reveal the demand for programming languages like VB.NET, Bash, and Python, alongside specialized languages including Dart, Kotlin, and Ruby, reflecting the growing importance of niche expertise. IT professionals are expected to be proficient in tools such as MS Visual Studio, and Crystal Report, as well as emerging frameworks like Flutter while skills in NoSQL and tools like MS Excel, ERP platforms, and Vertica are vital for supporting data-driven decision-making and business intelligence. By offering actionable insights into evolving industry expectations, this research serves as a valuable resource for educators with curriculum development, for recruiters with their talent acquisition strategies, and for job seekers with acquiring in-demand skills, ultimately ensuring a more responsive and competitive IT labor market.
Keywords:	Generative AI, Keyword Extraction, IT Employment, Large Language Models (LLMs), Natural Language Processing (NLP), Data Visualization, Data Analytics, Text Mining
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	DERMXNET: A HYBRID DEEP LEARNING AND GRADIENT BOOSTING APPROACH FOR EFFICIENT SKIN DISEASE DETECTION
Author:	N ANNALAKSHMI , S UMARANI
Abstract:	Skin diseases, including melanoma and non-melanoma, are among the most prevalent health concerns worldwide. Early and accurate detection of these conditions is critical for effective treatment and improved patient outcomes. This paper introduces DermXNet, a hybrid model combining Artificial Neural Networks (ANN) and eXtreme Gradient Boosting (XGBoost) for accurate classification of skin conditions, including melanoma, non-melanoma, and normal skin. The methodology involves systematic data pre-processing techniques such as resizing, normalization, and artifacts removal. The ANN module extracts high-dimensional features, which classified by XGBoost, leveraging the strengths of both deep learning and gradient boosting. The proposed model evaluated against nine existing deep learning models, including ResNet50, VGG16, and EfficientNetB0. DermXNet achieved the highest accuracy of 98.38%, surpassing other models in metrics such as precision, recall, F1-score, and AUC-ROC. Additionally, DermXNet demonstrated computational efficiency with a training time of 95 seconds and a model complexity of just 4 million parameters, making it suitable for real-world deployment. The results underscore the effectiveness of hybrid architectures in medical diagnostics. Future research can extend DermXNet to multi-class classification and integrate advanced domain-specific features to enhance its applicability.
Keywords:	Skin Disease Detection, Deep Learning, Artificial Neural Network, XGBoost, Hybrid Model, Medical Imaging, Melanoma Classification, Feature Extraction
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	UGECGAD: EVALUATING THE EFFICACY OF UNSUPERVISED GAN ARCHITECTURES FOR ECG ANOMALY DETECTION
Author:	SHAIK JANBHASHA, NARASIMHA RAO THOTA, PRASHANT ATMAKURI, G. ASHOK KUMAR, ATTRU HANUMANTHARAO, NEERUKATTU MALLIKHARJUNA RAO
Abstract:	Anomaly detection in Electrocardiogram (ECG) signals is critical for reliable diagnosis and continuous patient monitoring. Although traditional approaches are incapable of describing the rich nonlinear structure in ECG data, they are ineffective in detecting these abnormalities. Although Generative Adversarial Networks (GANs) has recently attracted increasing interest for ECG anomaly detection, research available currently does not provide a systematic evaluation of their advantages and disadvantages for practical situations. This paper addresses this gap by critically reviewing unsupervised GAN-based approaches, evaluating their ability to reconstruct normal ECG signals and accurately detect deviations. We introduce a rigorous empirical comparison of different GAN architectures and their adversarial variants, highlighting key challenges in their implementation. Our results show that the ECG-Adversarial Autoencoder (ECG-AAE) is superior to the other GAN-based approaches in terms of training anomaly, and provides the best performance in anomaly detection. This study contributes new insights into the robustness of ECG-AAE, establishing its potential for precise and reliable ECG anomaly detection in practical healthcare applications.
Keywords:	Generative Adversarial Networks, Deep Learning, Electrocardiogram, Anomaly Detection, Unsupervised Learning
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	USING ARTIFICIAL INTELLIGENCE SOFTWARE FOR CORRECTING INCORRECT INTONATION IN SINGING
Author:	XIAO HUIWEN
Abstract:	The purity of intonation has an impact on the expressiveness of vocal performance. The aim of this study is to determine the features of using artificial intelligence (AI) software for correcting incorrect intonation in singing. The research employed the following methods: analysis, comparison, calculations of two-factor analysis of variance (ANOVA), Cognitive Change Index (CCI). Training techniques included the use of the iZotope RX 10 application for the purpose of flexibility of performance based on the development of singing breathing. The Clip.audio application was used to work on ways of reproducing sounds; Adobe Audition – for creating a musical style. It was established that the advantages of digital applications are related to the possibility of automatic error detection and the development of individual vocal approaches that are important for students. The results showed that the learning process primarily contributed to the preservation of the correct intonation at different pitches, the development of correct pronunciation. The students developed a high level of vocal intonation during training, which affected the expressiveness of vocal performance. The practical significance of the article is aimed at the use of AI technologies for the development of vocal intonation. Artificial intelligence applications helped respondents achieve high vocal results, which also had a positive impact on the formation of vocal skills. Further research may focus on determining the ways of developing vocal intonation depending on the features of the musical genre.
Keywords:	Vocal Performance, Voice Mobility, Musical Articulation, Sound Image, Interactive Technologies
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	DEGREE OF IMBALANCE FOR TASK SCHEDULING IN CLOUD COMPUTING
Author:	NORA OMRAN ALKAAM, ABU BAKAR MD. SULTAN, MASNIDA HUSSIN, KHAIRONI YATIM SHARIF
Abstract:	The increasing reliance on cloud computing platforms for data-intensive applications has made efficient task scheduling a critical factor for maximizing resource utilization and minimizing the degree of imbalance. Inefficient task scheduling in these complex, distributed environments can lead to significant performance bottlenecks, including workload imbalance, which negatively impacts overall system efficiency and scalability. Task scheduling in cloud environments is a well-known NP-complete problem, further compounding the challenge of achieving optimal solutions. While existing metaheuristic approaches offer some mitigation, they often struggle to effectively balance exploration and exploitation, leading to suboptimal solutions and slow convergence. This study addresses this critical need by refining the previously proposed Henry Gas-Harris Hawks Modified Opposition (HGHHM) algorithm to explicitly minimize the degree of imbalance in task scheduling. We introduce a novel integration of Henry Gas Solubility Optimization (HGSO) and Harris Hawks Optimization (HHO) enhanced with a Modified Comprehensive Opposition-Based Learning (MOBL) strategy. This unique combination allows HGHHM to effectively explore the solution space while exploiting promising regions, leading to a more balanced workload distribution. Simulations using the CloudSim toolkit demonstrate that the improved HGHHM algorithm significantly reduces the degree of imbalance compared to the Cuckoo-based Discrete Symbiotic Organism Search (CDSOS) technique, achieving superior performance in terms of convergence speed and solution quality while avoiding local optima. A t-test confirms the statistical significance of these improvements, highlighting the potential of hybrid metaheuristic methods for optimizing task scheduling in large-scale cloud computing environments.
Keywords:	HGHHM; Cloud computing; Meta-heuristic; Scheduling; Optimization; Degree of imbalance.
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	AUTOMATIC SEGMENTATION OF SPONDYLOLISTHESIS AND SCOLIOSIS OF X-RAY IMAGES USING LIGHT-WEIGHT RESUNET ARCHITECTURE
Author:	SHIVA PRASAD K , Dr. JANA S
Abstract:	Scoliosis is the term for a curvature of the lumbar or thoracic spine in the coronal plane. Young children may develop spondylolisthesis and scoliosis, which, if left untreated, can grow into terrible agony. Lung and heart problems can also be caused by severe scoliosis. Early diagnosis can therefore help to arrest the disease's progression and make it easier to apply therapies or interventions. Recent research has focused on the use of convolutional neural networks for the diagnosis of scoliosis and spondylolisthesis on X-ray images. Unfortunately, the majority of the current approaches ignore the larger-scale image contextual feature information in favour of gathering feature information for prediction from localised parts of images. Furthermore, important features for classification—such as co-occurrence connections between labels and anatomical segmentation knowledge—are not completely utilised. This research suggests a Light-weight ResUNet (LW-RUnet) architecture that leverages the Xception backbone for feature extraction in order to segment scoliosis and spondylolisthesis using X-ray images. While the ResUNet decoder uses high-level features, the middle decoder uses middle features to get spatial information. To fine-tune the disease area, the suggested architecture integrates the ResUNet decoder characteristics with the proposed middle decoder features. The outcomes of the suggested segmentation technique are verified using a number of measures, including accuracy, sensitivity, IOU, and dice similarity coefficient. With regard to the normal, scoliosis, and spondylolisthesis classes, the suggested method's segmentation accuracy is 99.28%, 98.25, and 98.34, respectively.
Keywords:	Deep Learning, Scoliosis, Spondylolisthesis, Radiology, and X-ray images
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	THE EFFECTIVENESS OF DIGITAL COMMUNICATIONS FOR PROMOTING BRANDS ON SOCIAL MEDIA PLATFORMS
Author:	DIANA FAYVISHENKO, SVITLANA KOVALCHUK, DANYLO SIDIELNIKOV, OKSANA HOLIK, IVAN KYIANYTSIA
Abstract:	The effectiveness of digital communications is a key factor in the successful promotion of brands on social media platforms. This issue is especially relevant in the context of the rapid development of digital technologies, which create new opportunities for interaction with the audience and strengthening brand positions. However, despite extensive research on social media marketing, there remains a knowledge gap regarding the comparative effectiveness of different platforms and their content strategies in driving audience engagement and conversions. Existing studies often focus on individual platforms or specific digital marketing techniques without a comprehensive cross-platform analysis of digital communication effectiveness. This study addresses this gap by analyzing the effectiveness of digital strategies in promoting brands on Facebook, Instagram, TikTok, Twitter, and LinkedIn. Special attention is paid to key performance indicators such as the Engagement Rate (ER), Click-Through Rate (CTR), and conversions, as well as their dependence on content strategies and platform specifics. The aim of the study is to assess the impact of digital communications on the effectiveness of brand promotion in social media and identify key factors that contribute to increasing the rate of interaction with the audience. The study employs quantitative analysis, sociological surveys, comparative analysis, and correlation analysis to provide a data-driven evaluation of digital marketing strategies. The results demonstrate that TikTok achieves the highest ER due to interactive content and active engagement from a younger audience, while Instagram remains dominant in the premium segment due to its visually appealing content. Facebook maintains a stable engagement rate due to its broad user base but lacks innovative content formats, and LinkedIn proves effective for professional communication. Conversely, Twitter faces challenges due to declining user activity and engagement. The academic novelty of this research lies in its comprehensive cross-platform analysis of digital communication effectiveness, considering algorithmic operations, audience behavior, and content strategies. This study contributes to the existing body of knowledge by providing insights into the optimal digital marketing strategies across different platforms and industries. Prospects for future research include analyzing the adaptation of emerging content formats to platform-specific requirements, assessing the role of micro-influencers in audience engagement, and forecasting the evolution of digital communications until 2028.
Keywords:	Digital Communications, Social Media, Branded Content, TikTok, Instagram, Interaction, KPI, Conversions.
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	USING AI TO COUNTER CRIMINAL OFFENCES AGAINST THE WILL, HONOUR, AND DIGNITY OF A PERSON
Author:	VIACHESLAV SHEVCHENKO , OLEH ZVONAROV , LEONID SHCHERBYNA , YANA LUTSENKO, VOLODYMYR PYVOVAROV
Abstract:	In the era of the digitalization of society, crimes against the will, honour, and dignity of a person, in particular cyberbullying, blackmail, and threats, are becoming increasingly common. These phenomena are becoming increasingly common due to the anonymity of online communication and the rapid dissemination of information. This creates new challenges for law enforcement agencies, requiring the implementation of innovative technologies, in particular artificial intelligence (AI), for their detection, prevention, and investigation. Artificial intelligence (AI) opens up opportunities for automated analysis of threatening messages and prediction of risks of aggressive behaviour. The research focuses on assessing the potential of AI in combating such crimes, analysing its effectiveness, and identifying the limitations of technology use in law enforcement practice. The study evaluates the effectiveness of AI algorithms for detecting cyber threats and identifies their key limitations. The aim of the study is to assess the effectiveness of AI algorithms for automated detection of threatening messages, prediction of risks of criminal behaviour compared to traditional methods. The study employed the following methods: text data analysis using natural language processing (NLP), behavioural pattern modelling to predict risks, and surveys of human rights defenders and lawyers to study their attitudes towards AI in combating crime. The methods used include text data analysis, behavioural pattern modelling, and a sociological survey of human rights activists regarding their attitudes towards AI. The results showed that AI algorithms demonstrate high accuracy rates in detecting cyber threats, outperforming traditional methods in terms of speed and scalability. NLP algorithms achieved 85% accuracy compared to 75% in manual moderator analysis, confirming their effectiveness. At the same time, a survey of specialists revealed a number of ethical and legal limitations in the implementation of these technologies, in particular, 60% of respondents indicated the need for strict regulation of AI, and 35% emphasized the risk of false accusations. The academic novelty of the study is the interdisciplinary approach that integrates technological analysis with legal aspects to assess the effectiveness of AI, but also to identify obstacles to its practical application. The novelty of the study lies in combining the analysis of the effectiveness of AI with an assessment of the possibilities of its actual use in the fight against cyber threats. Further research should focus on adapting algorithms to the changing conditions of the digital environment and developing regulatory mechanisms for their implementation in compliance with human rights. Further research should improve AI algorithms and reduce the false positive rate.
Keywords:	Artificial Intelligence, Criminal Offences, Cyber Threats, Legal Ethics, Text Analysis.
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	BLOCK CHAIN-INTEGRATED MULTI-FACTOR AUTHENTICATION SYSTEM FOR ENHANCED SECURITY AND DECENTRALIZED INTEGRITY IN ONLINE EXAMINATION PLATFORMS
Author:	VALLEM RANADHEER REDDY, SHANKAR LINGAM GOURISHETTY, AMGOTH ASHOK KUMAR, PEESALA ILANNA
Abstract:	With the increasing depended on online examination systems, so providing security, transparency, and scalability has become a significant challenge.While existing methods with AI and deep learning-based multi-factor authentication provide robustness, they may not address data integrity and traceability effectively. In this research, we propose a novel framework that integrates Blockchain technology with multi-factor authentication system to enhance security and decentralization in online examinations. By combining Blockchain’smethod with advanced facial recognition and behavioral biometrics, the system ensures good data management and user authentication. The system uses a Hybrid Deep Learning (HDL) model for biometric verification and Blockchain smart contracts for secure transaction processing and audit trails. Experimental results demonstrate that the proposed system achieves 99.1% accuracy in user authentication while providing end-to-end security against unauthorized access and data manipulation.
Keywords:	Blockchain, Multi Factor Authentication, Deep Learning, Online Examination.
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	HOT TOPICS DETECTION AND PERFORMANCE OPTIMIZATION FROM MICROBLOGS USING HYBRID HADOOP FRAMEWORK
Author:	Dr. D. DAVID NEELS PONKUMAR, Dr. S. RAMESH, Dr. R. KALPANA, K. MANIKANDAN, N. RAGHAVENDRAN, Y. HAROLD ROBINSON, Dr. G. UMA MAHESWARI, Dr. SANDRA JOHNSON, Dr. B. PRATHUSHA LAXMI, R. SARAVANAKUMAR
Abstract:	The Hadoop framework's adoption is on the rise. We try to improve Hadoop's performance by incorporating a more sophisticated framework into the MapReduce paradigm, all while keeping the native Hadoop Framework's characteristics intact. The improved Hadoop framework sorts a large collection of microblogs according to the amount of attention they received from the social media site during a certain period 't'. There is a 3% decrease in the execution time of microblogs that have attained the attention level compared to those that have not. By distributing the load evenly, the EHF speeds up the processing of microblogs that have garnered a lot of interest. Global social media users may dynamically develop insoluble information. Social media networks employ big data to manage their massive data. Hadoop-based cloud platform provides large data fault tolerance and dependability. The foundation of big data analytics is Hadoop. The main drawback of Hadoop is processing massive configuration metrics. This paper proposes the Hybrid Hadoop Framework to improve big data processing by balancing workload, response time, network bandwidth, and hot topic detection for microblogs using cloud-based Apache Spark. To accurately find hot topics in large datasets, we purposely build MapReduce tasks. Experimental findings show that the suggested system is more accurate than comparable systems.
Keywords:	Hadoop Framework, Social Media, Mapreduce, Big Data Analytics, Apache Spark, Cloud, Workload, Response Time, Network Bandwidth, Hot Topic Detection
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	MAXIMIZING THE EFFICIENCY OF AI BASED IOT DEVICES WITH UPGRADED 802.11 AND 802.16 BASED NETWORKS TO OPTIMIZE NETWORK CONGESTIONS
Author:	KOUSHIK REDDY CHAGANTI, DR.Y. RAMAMOHAN REDDY, DR.Y. ALEKYA RANI, D. SANDHYA RANI, PALLABOTHULA RAMESH, NANNAPARAJU VASUDHA
Abstract:	The proliferation of Internet of Things (IoT) devices has led to unprecedented data generation and transmission, demanding efficient networking solutions. This paper proposes a novel approach to maximize the efficiency of AI-based IoT devices by leveraging upgraded 802.11 and 802.16 networks for effective congestion management. Traditional networking protocols often struggle to cope with the dynamic and data-intensive nature of IoT applications, leading to network congestion and performance degradation. This study provides an in-depth exploration of interference in wireless applications, with a particular focus on the distinctive interference challenges posed by 5G and IoT. It elucidates various optimization techniques aimed at mitigating these challenges. Emphasizing the criticality of managing interference and optimizing network performance in 5G environments, the paper underscores their pivotal role in facilitating dependable and efficient connectivity for IoT devices. Such connectivity is indispensable for ensuring the smooth operation of business processes.To address this challenge, we integrate advanced artificial intelligence (AI) algorithms with enhanced 802.11 (Wi-Fi) and 802.16 (WiMAX) standards to optimize network resource utilization and mitigate congestion. Our proposed solution dynamically adapts to changing network conditions, intelligently prioritizes traffic, and optimally allocates resources to ensure seamless connectivity and superior performance for AI-driven IoT devices. hrough extensive simulations and practical validations, we demonstrate the efficacy of our approach in significantly reducing latency, improving throughput, and maximizing overall network efficiency, thus empowering AI-based IoT ecosystems with enhanced connectivity and performance.
Keywords:	AI-Driven Iot Devices, Efficiency Enhancement, Advanced Networking, Congestion Alleviation, Traffic Prioritization And Latency Reduction
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	EXPLORING THE ADVANCEMENTS AND CHALLENGES OF OBJECT DETECTION IN VIDEO SURVEILLANCE THROUGH DEEP LEARNING: A SYSTEMATIC LITERATURE REVIEW AND OUTLOOK
Author:	M.KOTESWARA RAO , P.M. ASHOK KUMAR
Abstract:	Video surveillance has become increasingly important in recent years, leading to a growing need for effective and accurate surveillance systems that can detect objects and understand scenes. To address these challenges, methods based on deep learning have emerged as the state-of-the-art approach and have shown remarkable results in various applications. In this systematic literature review (SLR) paper, we investigate the recent advancements and challenges of object detection in video surveillance through deep learning. Our primary goal aim to present a comprehensive overview of the recent research developments in this field and to emphasize the challenges that should be addressed in future research. In this study, we analyzed various deep learning-based object detection methods and evaluated their performances based on several performance metrics. Our findings indicate that deep learning-based methods have demonstrated promising results in regards of accuracy and real-time performance in video surveillance for object detection. However, the study also highlights several challenges such as scalability, robustness, and interpretability that require further research. Finally, this SLR paper concludes with a discussion of future research directions in this field and offers a roadmap for future work. Our study results can serve as a useful reference for researchers and practitioners working in the field of video surveillance and deep learning.
Keywords:	Object detection, Video surveillance, Deep learning, Object tracking, Scene understanding, Performance analysis
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	CYBER FORTIFICATIONS: ENSEMBLE SECURITY THROUGH DEEP LEARNING INNOVATIONS
Author:	Y. SUDHEER KUMAR , DR. A MARY SOWJANYA
Abstract:	Cyber threats using traditional approaches are more complex to detect in the early stages. Cybersecurity is the domain that provides appropriate defense against modern attacks. Cyber threats or cyber-attacks are attempts to destroy, thieve, alter, disable, or destroy data or applications using unauthorized access to a network, computer system, or digital device. Detection and classification of cyber attacks is a tedious task for state-of-art algorithms because of its more computation time and lack of understanding of attack patterns. Deep Learning (DL) is a domain used in many cyber threat detection systems, such as distributed denial-of-service (DDoS) attacks, phishing, ransomware, and anomalies. The pre-trained model Attack Pattern Convolutional Neural Networks (AP-CNNs) is introduced to detect attack patterns from the cyber attacks dataset. This paper introduces an Ensemble Security Model (ESM) to detect and classify cyber attacks from benchmark datasets. ESM combines Deep Neural Network (DNN) and Adaptive Support Vector Machine (ASVM). DNN is used as a vital feature extraction that extracts significant attack patterns. ASVM is used to classify different types of attacks obtained from other datasets. This paper uses two benchmark datasets to measure the strength of the ESM. The pre-trained model Transfer Learning with a BERT-based Model is used to train on given datasets. Finally, the results show that the proposed approach obtained better results than existing models.
Keywords:	Distributed Denial-of-Service (DDOS) Attacks, Phishing, Ransomware, and Anomalies, Ensemble Security Model (ESM), Deep Neural Network (DNN).
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	EFFICIENT SUPERVISED MACHINE LEARNING FOR CYBERSECURITY APPLICATIONS USING ADAPTIVE FEATURE SELECTION AND EXPLAINABLE AI SCENARIOS
Author:	DR. NAGA ANURADHA, M. SAILAJA, DR. PRABHAKAR MARRY, DR. DASARI MANENDRA SAI, PALLABOTHULA RAMESH, DR. SIVANANDA LAHARI REDDY
Abstract:	Cyber threats are changing continuously, so the need for robust, efficient, and adaptive machine learning solutions is of utmost importance for the protection of critical infrastructures. The current supervised machine learning models face issues in dealing with high-dimensional data, real-time adaptability, and privacy in data while delivering accurate and interpretable results. We present a multi-faceted framework that combines state-of-the-art techniques to optimize cybersecurity applications. We begin with a method called Adaptive Feature Selection Using Reinforcement Learning (AFSRL), which dynamically identifies the optimal feature subsets for the best classification, computational efficiency, and detection latency. This reduces dimensionality by 40–60% and improves model accuracy by 10–15%. We propose a Hybrid Ensemble Learning with Dynamic Weight Adjustment, which dynamically integrates these diverse algorithms: Random Forest, SVM, and Gradient Boosting. It obtains an 8-12% accuracy improvement together with a reduction of 15% in false positives. For complex attack patterns, CAGN exploits temporal and spatial relationships in network graphs for 20% improved precision while maintaining detection latency below 100ms. Our PPFL framework preserves privacy by protecting sensitive data while retaining model performance parity with centralized approaches. Finally, XAI-TADM brings trust and interpretability with SHAP and LIME, explaining actionable insights which improves response time by 25%. Such an all-round accurate, efficient, adaptive, privacy-sensitive, and transparent framework would work very well with real-time and high-stake environments in applications such as health care, finance, and national security systems.
Keywords:	Cybersecurity, Adaptive Feature Selection, Explainable AI, Federated Learning, Graph Neural Networks
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	EFFICIENT MRI SEGMENTATION OF SPINE HEMANGIOMAS: A NOVEL MODIFIED U-NET APPROACH TO ENHANCE TUMOR BOUNDARY DETECTION
Author:	ABUJALAMBO MAHMOUD I M , NOR AZLINAH BINTI MD LAZAM , MALIK JAWARNEH , SHADI M S HILLES , ABDALLAH ALTRAD
Abstract:	Magnetic Resonance Imaging (MRI) is a widely used, non-invasive method for medical imaging, particularly effective in visualizing soft tissues and identifying abnormalities like spine hemangiomas. One of the main challenges remains the low segmentation accuracy of skeletal MRI images. Spine hemangioma segmentation involves algorithmically identifying and localizing these tumors within MRI scans, a process crucial for accurate diagnosis and treatment planning. Although several segmentation methods exist, this paper introduces a U-Net-based approach, implemented in PyTorch and optimized with the Adam optimizer. This setup refines model weight adjustments and harnesses the full capabilities of a fully connected convolutional neural network (CNN) for precise semantic segmentation, including pixel-wise classification through an encoder-decoder structure. This U-Net architecture is versatile and adaptable to various analytical tasks across diverse applications. The model was trained on a substantial dataset spanning the three primary anatomical planes used in medical imaging—Axial, Coronal, and Sagittal without additional data augmentation. It achieved real-time segmentation with a remarkable accuracy of 94.13% and demonstrated strong performance metrics, including a Dice coefficient of 0.634 and Precision of 0.711, underscoring its robustness and potential clinical utility. This work highlights U-Net’s effectiveness in spine hemangioma segmentation and explores its matching capabilities, indicating promising potential for advancements in automated MRI analysis.
Keywords:	MRI Spine Hemangioma Segmentation, U-Net Model, Convolutional Neural Network (CNN), Semantic Segmentation, Precision, Dice coefficient, accuracy.
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	BUILDING ROBUST IOT NETWORKS WITH DYNAMIC LAYER PRIORITIZATION AND PREDICTIVE FAULT MANAGEMENT PROCESS
Author:	DASARI ANUSHA, M. GAYATHRI, B V CHOWDARY, 4DR. NARASIMHA CHARY CH, DR. ATHMAKURI SATISH KUMAR, DR. C NAGESH
Abstract:	The rapid proliferation of IoT networks places strict demands on the communication frameworks to be efficient, scalable, and robust for dealing with varied resource constraints. The existing architectures of IoT often bottle up in issues related to high energy consumption, failure in reliability, and issues related to interaction and complexity primarily in dynamic large-scale deployments. This hampers the real-time performance and overall reliability needed in applications such as healthcare systems, industrial automation, and smart grids. With respect to the previous, we aim for improving such challenges through using five novel techniques to optimize communication among machines as proposed by this work, using the approach: Dynamic Layer Prioritization, Context-Aware Energy Optimization, Adaptive Protocol Selection, Predictive Fault Management, and Hierarchical Data Aggregation. Dynamic layer reallocation with resource optimization in using reinforcement learning brings 30% to 40% reduction of packet loss with latency improved between 20% to 25%. CAEO relies on edge intelligence and context-aware sleep-wake cycles to boost battery life by 50% while retaining 95% data accuracy. APS uses machine learning to select the best protocols for communication, thereby increasing throughput by 20–30%. PFM makes use of predictive analytics and blockchain integration to prevent faults before they happen, thus enhancing network reliability by 25%. HDA reduces redundancy in data, which in turn reduces the overhead of data transmission by 40% and increases processing speed by 30%. This multi-layered approach ensures resource efficiency, real-time performance, scalability, reliability, security, and interoperability with diverse IoT ecosystems. The proposed model shows great impacts, including reduced operational costs, enhanced energy efficiency, and robust fault tolerance, making it a transformative solution for next-generation IoT networks.
Keywords:	IoT Networks, Dynamic Layer Prioritization, Predictive Fault Management, Energy Optimization, Adaptive Protocols
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	MORPHEME MASTERY FOR TAMIL QA: HARNESSING THE WISDOM OF NANNŪL FOR ACCURATE ANSWERS
Author:	NIVEDITHA S , PAAVAI ANAND G
Abstract:	Morpheme Mastery represents a groundbreaking advancement in Tamil Question Answering, introducing a pioneering answer alignment technique that leverages Finite State Transducer (FST) technology to expertly handle complex morphological structures, particularly nouns. This research contributes to the advancement of IT in natural language processing by developing a novel approach that ensures precise analysis and interpretation, enabling efficient and accurate processing. The significance of this research lies in its potential to revolutionize computational linguistics and Tamil Question Answering, setting a new benchmark for accuracy and efficiency. By incorporating novel steps such as stop words removal, Numeric Information Extraction (NIE), and Temporal Entity Recognition (TER), Morpheme Mastery addresses a critical gap in existing question answering systems. Our deliberate choice of a rule-based approach, rather than machine learning alternatives, guarantees consistently high accuracy, a critical prerequisite for building a robust computational language framework. Empirical results demonstrate the efficacy of our approach, yielding an impressive 11.87% improvement in CHAII and 9.58% in SQuAD dataset efficiency. This significant enhancement underscores Morpheme Mastery's potential to transform the landscape of Tamil Question Answering and computational linguistics, enabling more accurate and efficient human-machine interaction.
Keywords:	Answer Alignment, Tamil, Question Answering, Morpheme Analysis, Nannūl
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	DESIGN AND DEVELOPMENT OF MICROSERVICES-BASED CRM SYSTEM
Author:	ANDRII PRYHODA , ROSTYSLAV SIKORA , VOLODYMYR MOSKALENKO , ANDRII ROSKLADKA
Abstract:	The aim of the article is to analyse the technical and organizational aspects of the implementation of microservice architecture in CRM systems. The research employed analytical and logical methods, technical analysis of microservice architecture comparing it with monolithic systems, as well as Docker and Kubernetes applied technologies. The use of Docker and Kubernetes containerization tools is shown to be a key factor in successfully managing a microservices-based CRM infrastructure. The analysis of implementation results shows how the microservice architecture reduces operational costs and increase the efficiency of system support. The role of automation of information systems (IS) in business was analysed. It was established that the modern market creates a situation where it is necessary to constantly increase production efficiency. General requirements for the CRM system were formulated, including security and protection, reliability and availability, user-friendly interface, ease of implementation and maintenance, availability of basic functionality. Current web application development technologies are explored and the technologies used in the project, including Django, Django Rest Framework, React, MobX, and Ant Design are described.
Keywords:	CRM system, Docker, Kubernetes, Microservice architecture, Monolithic architecture.
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	A HYBRID FRAMEWORK FOR OBJECT DETECTION AND SEGMENTATION IN AUTONOMOUS VEHICLES USING YOLO NAS AND MASK R-CNN
Author:	DARTHY RABECKA V ,BRITTO PARI J
Abstract:	A leading opinion prescription that can precisely establish and segregate objects in complex environments is necessary for the hurried development of self-driving autos. The new hybrid framework shown in this work improves object detection and segmentation performance by combining Mask R-CNN with You Only Look Once Neural Architecture Search (YOLO NAS). With the neck and head of YOLO-NAS retained, this study tries to boost the performance of YOLO-NAS by substituting a combination of Res Net and Feature Pyramid Network (FPN) for the default Rep Ne X t backbone. Additionally, to increase segmentation capabilities, the study integrates Mask R-CNN. Our methodology leverages the efficiency of YOLO NAS for fast object detection and the precision of Mask R-CNN for complex segmentation tasks using the KITTI dataset, a leading benchmark in autonomous driving research. This approach resolves issues such as disparate object sizes, obstructions, and complex backgrounds that are frequently encountered when driving in urban areas. Our cloud-based approach outperforms previous approaches in terms of precision, recall, and F1 scores. The results of our experiments show that this combination approach could greatly contribute to the development of more reliable and safer autonomous driving systems, paving the way for advancements in real-time perception technology.
Keywords:	Convolution Neural Network, Object Detection, Autonomous Vehicles, You Only Look Once
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	AI BASED MODIFIED APPROACH IN HIGH-RESOLUTION IMAGE RESTORATION TECHNIQUES FOR SMART HEALTHCARE SYSTEM: A TECHNICAL REVIEW AND ANALYSIS
Author:	PRASANTA KUMAR SAHOO , DEBASIS GOUNTIA , RANJAN KUMAR DASH , MANAS KUMAR NANDA
Abstract:	Medical image analysis is the most prominent research area of digital image processing. In digital image processing, medical images are more sensitive and informative. Data loss in medical image processing harms patients and the healthcare system. Therefore, several medical image analysis techniques play a significant role in ensuring lossless information during disease diagnosis. The visual clarity of an image depends upon its high resolution. Therefore, high resolution is crucial in medical image analysis. The high-resolution problem occurs due to spatial resolution limitations. These limitations can be attributed to hardware limitations, low radioactivity doses, or the acquisition time of a specific image. To avoid these limitations, several high-resolution image restoration (or reformation) techniques are available. These techniques include image compression, histogram equalization, edge detection, feature extraction, image synthesis, and noise reduction, among others. Recently, different AI-based techniques, like machine learning (ML) and deep learning (DL), have been considered booming technologies for high-resolution image reformation. Therefore, we felt it was necessary to review various AI learning-based techniques and their applications in the reformation of high-resolution digital images. The digital images may include various types, such as medical images or those from the Indian Historical Society. In this article, we first shed light on different modalities of medical image analysis techniques with their image acquisition properties. We then apply machine learning and deep learning AI approaches to the image acquisition properties to address the issue of poor visual quality in the images. This serves as evidence that ML and DL models must modify the component configuration of medical image modalities to reform high-resolution images. To improve the quality of low-resolution images, we are building new models for histogram equalization, compression, decompression, and noise reduction based on this comprehensive review.
Keywords:	High-Resolution, Image Enhancement, Medical Image Analysis, Image Compression, Machine Learning, Deep Learning.
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	DEEP LEARNING-BASED CLASSIFICATION AND SEGMENTATION OF CHEST PATHOLOGIES
Author:	JAKKULA SRAVANTHI, BANDA VENKATA RAMANA, JUTHUKA ARUNA DEVI, SURAGALI CHANTI, SRADHANJALI PATTANAIK, PALTHIYA ANANTHA RAO, MARADA SRINIVASA RAO
Abstract:	Chest diseases, such as COVID-19, viral pneumonia, and lung opacity, in most severe cases, call for quick diagnosis and accurate treatment. Applying deep learning methods, mainly convolutional neural networks (CNNs), has become attractive among machine learning techniques for automated image diagnostics. This paper reports a new ensemble approach that utilizes CNN-1, CNN-3 and VGG-16 structures to classify the disease and U-Net for segmenting the chest diseases in X-ray pictures considered from the Kaggle repository. Data augmentation is applied to original samples to increase the size and performance of a model. The segmentation procedure shows a high capability to define interested regions in the lung, which contributes to higher accuracy, performed comparison between the proposed segmentation method and existing methods. The following results obtained by the model are 96% in segmentation and 87.7% in classification- reflecting the preferred method's high accuracy. The proposed model was evaluated on performance metrics like F1-score, Precision, and Recall. Therefore, this result may lead the way for enhanced diagnostic accuracy and treatment decisions in clinical settings.
Keywords:	Viral Pneumonia, COVID-19, Lung Opacity, Deep Learning, Ensemble, CNN-1, CNN-3, VGG 16, Segmentation, U-Net.
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text

Title:	IMPLEMENTATION OF HYBRID DEEP LEARNING CNN MODEL FOR MULTISPECTRAL SATELLITE IMAGE CLASSIFICATION IN LAND CHANGE DETECTION
Author:	YERIK AFRIANTO SINGGALEN, STEPHEN APRIUS SUTRESNO, MUHAMAD NUR AGUS DASRA, RUBEN WILLIAM SETIAWAN
Abstract:	This study investigates implementing a Hybrid Deep Learning Convolutional Neural Network (CNN) model for classifying multispectral satellite imagery to detect land cover changes in Untung Jawa Island, Indonesia. The research addresses critical limitations in conventional image classification methods that struggle with capturing subtle terrain modifications and complex land cover transitions at accelerated rates. Our key contribution is the development of an innovative CNN architecture that integrates multiple deep-learning approaches optimized for different spectral bands, significantly enhancing feature extraction and classification capabilities. By systematically applying the Normalized Difference Vegetation Index (NDVI) and Normalized Difference Built-up Index (NDBI), the study demonstrates substantial improvements in classification accuracy, achieving rates exceeding 85% across multiple temporal datasets—a significant advancement over traditional methods that typically achieve 65-75% accuracy in similar contexts. The hybrid CNN model successfully processes over 1,000 image patches while maintaining consistent accuracy levels above 82% in feature extraction tasks. Quantitative analysis reveals a 28% increase in urbanized areas between 2013 and 2024 and a 19% decrease in vegetated surfaces, providing crucial evidence for environmental planning. Implementing U-Net architecture for image segmentation further enhances the model's capability to detect subtle environmental modifications, particularly in coastal regions where rapid urbanization intersects with sensitive ecological systems. This research advances remote sensing technology by establishing new methodological benchmarks for automated environmental monitoring systems and providing actionable insights for sustainable urban development planning in vulnerable small island ecosystems.
Keywords:	Hybrid Deep Learning, Convolutional Neural Network, Multispectral Satellite Image, Land Change Detection, NDVI, NDBI, U-Net Architecture, Environmental Monitoring
Source:	Journal of Theoretical and Applied Information Technology 31^st March 2025 -- Vol. 103. No. 6-- 2025
Full Text