		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
January 2025 | Vol. 102 No.1

Title:	DESIGN OF AN IMPROVED METHOD FOR INTRUSION DETECTION USING CNN, LSTM, AND BLOCK CHAIN
Author:	D.JYOTHI , P.JAMES VIJAY , M.KISHORE KUMAR , Dr.R.VENKATA LAKSHMI
Abstract:	The sophistication of cyber-attacks has brought the need for proactive Network Intrusion Detection Systems that will handle real-time threats and maintain data integrity and privacy. The traditional signature-based intrusion detection solutions have a number of drawbacks. These limitations consist of high false positives, non-transparency of decision-making, and vulnerability to tampering. Besides this fact, a centralized approach brings serious risks of privacy disclosure and raises scalability challenges, especially within distributed environments. This work proposes overcoming these limitations by integrating block chain technology along with state-of-the-art latest AI techniques to provide a powerful, decentralized, and explainable intrusion detection framework. This paper proposes a system integrating three key methodologies: (1) multimodal intrusion detection using CNN and LSTM networks for both spatial and temporal features of network traffic with 98.2% detection accuracy and less than 2.3% false positives. The decentralized model was trained using the adaptive federated learning with differential privacy in the following sections, which achieved 95.4% accuracy with 50% faster model convergence compared to the centralized approaches. Block chain is used to log updates of the federated model securely, thus guaranteeing tamper-proof auditability. SHAP is utilized for XAI, explaining AI-driven decisions in an understandable manner. 99.9% of decisions are explainable and will be immutably stored on the block chain for post-event forensic analysis. Such a fusing of the different methods described above will enable a highly accurate, scalable, and transparent NIDS. The new approach goes beyond mere security because early threat detection allows trust, accountability, and privacy-something comprehensive for modern decentralized networks.
Keywords:	Intrusion Detection, Block chain Security, Federated Learning, Convolutional Neural Networks, Explainable AI, and Process.
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	ANOMALY DETECTION IN NEXT-GEN IOT:GIANT TREVALLY OPTIMIZED LIGHTWEIGHT FORTIFIED ATTENTIONAL CONVOLUTIONAL NETWORK
Author:	Dr. K. SUJATHA, Dr.KALYANKUMAR DASARI, S. N. V. J. DEVI KOSURU, NAGIREDDI SURYA KALA, Dr. MAITHILI K, Dr.N.KRISHNAVENI
Abstract:	Within the most recent version of security monitoring solutions crafted for interconnected device networks faces limitations due to data scarcity, diverse device types, and limited computational resources. Unlike traditional solutions, these networks require a different approach. To address these limitations, the paper introduces LF-ACANet-GTOA, a novel approach leveraging a unique architecture called Lightweight Fortified Attentional Convolutional Network. This model is optimized with the Giant Trevally Optimization Algorithm (GTOA) for efficient and accurate intrusion detection within resource constrained IoT networks. The system focuses on critical information within network traffic data using an attention mechanism. It analyses two public datasets such as CIC-IDS-2017 and Bot-IoT to assess the effectiveness of LF-ACANet-GTOA. A meticulous pre-processing stage ensures clean and consistent data for the model. It provides a detailed description of the LF-ACANet-GTOA design, encompassing its components: Convolutional Encoder, Feature Enrichment Block, Attention Mechanism Integration, and Classification Layer. Additionally, it utilizes the Giant Trevally Optimization Algorithm (GTOA) for efficient training and optimization. The simulation results for the proposed LF-ACANet-GTOA method on the CIC-IDS-2017 dataset are promising, achieving high accuracy (99.57%), precision (99.26%), recall (99.16%), and F-score (99.21%), with low false alarm (0.73%) and miss rates (0.83%). These results suggest that LF-ACANet-GTOA has the potential to be a robust and secure solution for intrusion detection in resource-constrained interconnected device networks.
Keywords:	Security monitoring solutions, Interconnected device networks, Lightweight Fortified Attentional Convolutional Network (LF-ACANet-GTOA), Giant Trevally Optimization Algorithm (GTOA), CIC-IDS-2017 and Bot-IoT.
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	THE ROLE OF THE DIGITAL BUSINESS ECOSYSTEM IN INNOVATIVE AND INTELLECTUAL DEVELOPMENT OF REGIONS
Author:	OLHA POPELO, VIKTORIIA MARHASOVA, OLENA PEREPELIUKOVA, OLENA KAKHOVSKA, MYROSLAVA OPRYSOK, SERHII KHOMENKO
Abstract:	The article is devoted to the analysis of features of development of the digital entrepreneurial ecosystem in innovative and intellectual development of regions. The essence of the digital entrepreneurial ecosystem has been studied, and attention has been focused on three main elements, namely: digital citizenship of users, digital technology entrepreneurship and digital multilateral platform. Prerequisites contributing to formation of digital entrepreneurial ecosystems are outlined, as follows: progress of digital technologies; change in the competition nature; desire of customers to satisfy their needs online; urgent need to unite geographically separated economic subjects through globalization. Structured key factors affecting the innovative and intellectual development of the region in the context of the development of digital entrepreneurial ecosystems. The key performance indicators of development of digital entrepreneurial ecosystems are analyzed. Recommendations for further development of digital entrepreneurial ecosystems are provided.
Keywords:	Ecosystem, Entrepreneurship, Universities, Education System, Digital Entrepreneurial Ecosystem, Digital Transformation, Innovation, Innovative And Intellectual Development, Region, Public Authorities, State Policy
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	ENHANCING SHORT TEXT ANALYTICS THROUGH MULTIVARIATE FILTER METHODS FOR EFFICIENT FEATURE SELECTION
Author:	EMAD QAIS , VEENA MN
Abstract:	Feature selection is an essential process in text analytics that allows for the identification of the most pertinent elements while minimising random noise in the data. Multivariate filter methods provide robust methodology for feature selection by utilising statistical measures across several variables. The present research introduces a technique for brief text analytics that employs multivariate filter methods to pick features. The enormous volume of brief text data produced in many fields presents considerable difficulties in obtaining valuable insights from this data. A novel method is presented in this paper to improve brief text analysis by efficiently choosing the most pertinent characteristics through the use of multivariate filters. Through a sequence of meticulously planned experiments, the suggested model showcases enhanced precision, comprehensibility, and computational resource utilisation. The findings underscore notable progress in predictive metrics, suggesting that this methodology can have a large influence on the domain of text analytics. In addition to advancing natural language processing, this work provides practical applications in sentiment analysis, social media monitoring, and customer feedback interpretation. Given the ongoing evolution of digital communication, this approach is well-positioned to reveal subtle insights in concise textual material, which is frequently rich in important information.
Keywords:	Feature Extraction, Feature Selection, Text Analytics, Short Text, Multivariate filter Methods.
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	HYBRID DEEP LEARNING FRAMEWORK FOR INTRUSION DETECTION: INTEGRATING CNN, LSTM, AND ATTENTION MECHANISMS TO ENHANCE CYBERSECURITY
Author:	A. HARSHAVARDHAN DR. M. SREE VANI DR ANITHA PATIL NAGENDAR YAMSANI KANDE ARCHANA
Abstract:	Advanced intrusion detection systems (IDS) are required to protect contemporary networks due to the increasing complexity of cyber attacks. Due to their inability to fully capture the complex temporal and spatial patterns in network traffic, traditional intrusion detection systems (IDS) methods—such as standalone machine learning and deep learning models—frequently result in high false-positive rates and decreased detection accuracy. These drawbacks show how creative frameworks that can handle these problems are required. A hybrid deep learning framework that combines Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks and an attention mechanism is proposed in this study. The framework effectively captures spatial features, sequential dependencies, and critical network traffic patterns, enhancing accuracy and interpretability. The methodology includes comprehensive data preprocessing, principal component analysis (PCA) for dimensionality reduction, and recursive feature elimination (RFE) for feature selection. Hybrid Deep Learning-based Intrusion Detection (HDLID), a revolutionary algorithm, directs the suggested system's implementation. Tested on the UNSW-NB15 dataset, the framework outperforms state-of-the-art precision, recall, and F1-score models, achieving an impressive accuracy of 97.89%. The results validate its robustness and scalability for real-world applications. The proposed framework offers a practical, high-performance solution for intrusion detection, addressing limitations in existing methodologies and contributing to improved cybersecurity in diverse network environments.
Keywords:	Intrusion Detection System (IDS), Deep Learning, Cybersecurity, UNSW-NB15 Dataset, Network Threat Detection
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	RELIABLE MULTIHOP VORONOI LEACH FOR EFFICIENT WIRELESS SENSOR NETWORK
Author:	D SATYANARAYANA, ABDULLAH SAID Al KALBANI
Abstract:	Energy efficient processes are substantially important in Wireless Sensor Networks, especially in communication processes. In order to lower energy utilization on LEACH, Voronoi diagram-based clusters were presented that can bring a distance relationship among the given set of nodes. Multihop Voronoi LEACH was proposed to further reduce the communication energy expenditure between sensor nodes and cluster heads within Voronoi region. However, the method may often experience communication interruptions if the data to be transferred from sensor node to cluster head in high volumes or frequent data transfers. In this article, we propose an energy efficient wireless sensor network protocol called Reliable Multihop Voronoi LEACH, which is suitable for bulk data transfer in sensor network applications. Simulation results determine the performance of proposed method.
Keywords:	Sensors, Wireless Sensor Networks, Energy Saving, LEACH, Voronoi Diagrams.
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	EFFICIENT OBJECT DETECTION IN AGRICULTURAL ENVIRONMENTS IMPLEMENTING COLOR FEATURES EXTREME LEARNING MACHINE
Author:	DESIDI NARSIMHA REDDY, BALA BRAHMESWARA KADARU, A L SREENIVASULU, R. KANCHANA, PRADEEP JANGIR, CHERUKUPALLI RAMESH KUMAR
Abstract:	In a wide variety of computer vision applications, such as surveillance systems, autonomous cars, and environmental monitoring, object detection is an extremely important component. For the purpose of conducting effective analysis and making sound decisions, it is vital to have object identification methods that are both accurate and efficient in pastoral environments, which are characterized by the presence of animals and other things. The purpose of this research study is to present a unique method for the rapid recognition of objects in pastoral landscapes by utilizing a Color Feature Extreme Learning Machine (CF-ELM). For the purpose of achieving higher object detection performance while maintaining computational efficiency, the CF-ELM integrates color characteristics with the ELM algorithm. The proposed method is shown to be successful and efficient in detecting objects in pastoral environments, as demonstrated by the results of the experiments.
Keywords:	Object Detection, Pastoral Landscapes, Color Feature Extreme Learning Machine, and Color Features.
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	A NOVEL DEEP LEARNING METHOD FOR DETECTING CHANGES IN SATELLITE IMAGERY
Author:	Diaa Hafez Ibrahim, Reda A. El-Khoribi, Farid Ali MousaThe identification of change information has been crucial to the use of satellite imagery, the monitoring of land cover and use, the estimation of damage from natural catastrophes and the detection of military targets. There are numerous conventional techniques for detecting changes in multispectral remote sensing images, but they frequently fall short of our needs for durability, accuracy, and precision. This paper introduces a novel deep learning method for identifying changes in satellite data, with an emphasis on environmental changes and urban expansion. Using 24 pairs of Sentinel-2 satellite image data collected from 2015 to 2018, of which 10 pairs were used for testing and 14 pairs for training. Thirteen spectral bands with different spatial resolutions (10 m, 20 m, and 60 m) make up each multispectral image pair. The paper evaluates changes in urban and rural environments using the visible spectrum bands (2, 3, and 4) at a resolution of 10 m. The collection contains manually annotated changes. The paper compares the results obtained from the proposed solutions by Siamese Network and U-Net to address this change detection problem. With an accuracy of 0.86, the Siamese Network is used to detect high-level structural changes between pre- and post-event images by learning similarities between paired images. With an accuracy of 0.84, U-Net, which is intended for semantic segmentation, offers pixel-level predictions that improve change detection detail while the hybrid method for pixel-level change detection that combines the Siamese Network and U-Net in order to increase accuracy even further. This approach, which uses the Siamese Network for patch-wise similarity comparison and U-Net for fine-grained pixel segmentation, yields the maximum accuracy of 0.91. An efficient framework for applications in urban planning, crisis management, and environmental monitoring is suggested by the suggested hybrid technique, which shows great promise for accurate and thorough change detection in satellite imagery
Abstract:	Satellite Images, U-net, Siamese, Urban development, Change detection
Keywords:
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	NOVEL SYSTEM OF WIRELESS POWER TRANSFER FOR BATTERY CHARGING OF EV APPLICATIONS USING QUASI ZSI
Author:	Dr. J BHAVANI , JUPALLI PUSHPA KUMARI , Dr.T.NIREEKSHANA, PUNITHA KASU, AMULYA CHENNARAM
Abstract:	This research describes a 60-kW quasi-impedance source-based inductive type wireless power transfer quick charging system. The suggested system comprises of a quasi-Z-source converter (needed to raise the input voltage), a high-frequency inverter, and an inductive coil with capacitive compensation. At the receiver end, a buck converter is installed, which is operated in Constant Current (CC) or Constant Voltage (CV) charging modes based on the State of Charge. Two different PI controllers are used to achieve both CC and CV. The suggested system generates 60 kW of power at 135A, charging the battery from 0 to 80% SOC in less than 30 minutes. The simulation results achieved using MATLAB are consistent with the commercial guidelines for the battery charging process. Detailed simulation outcomes are provided, helping us to take use of many characteristics of the intended charger.
Keywords:	Quasi Z Source Converter, Wireless Power Transfer (WPT), Design Of Converter, Battery Charging.
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	LEVERAGING EXPLAINABLE AI TO IMPROVE BREAST CANCER DETECTION RATE USING TRANSFER LEARNING WITH DEEP RECURRENT CONVOLUTIONAL NEURAL NETWORKS
Author:	DR MOHAMMED ZAHEER AHMED, DR J VISUMATHI, DR PRAVEEN KUMAR YECHURI, DR A SWARNA, G SUREKHA, P. NARESH
Abstract:	Breast cancer remains a significant global health concern, emphasizing the need for accurate and timely diagnosis. Leveraging advancements in artificial intelligence (AI), particularly deep learning techniques, has shown promise in improving breast cancer detection rates. In this study, we propose a novel approach that integrates explainable AI principles with transfer learning using deep recurrent convolutional neural networks (RCNNs) to enhance breast cancer detection. The proposed model combines the spatial feature extraction capabilities of convolutional layers with the sequential processing capabilities of recurrent layers, thereby capturing both local patterns and temporal dependencies in mammogram images. Additionally, the incorporation of explainable AI techniques facilitates interpretation and understanding of the model's decisions, enhancing its clinical utility. We evaluate the performance of the proposed approach on a publicly available mammography dataset and demonstrate its effectiveness in improving breast cancer detection rates compared to baseline models. Furthermore, we provide insights into the learned representations and decision-making processes of the model, thereby enhancing transparency and trust in AI-assisted diagnosis. Our findings underscore the potential of explainable AI-driven transfer learning with deep RCNNs as a valuable tool for augmenting radiologists' capabilities and improving patient outcomes in breast cancer screening and diagnosis.
Keywords:	Breast Cancer Detection, Deep Learning, Deep Recurrent Convolutional Neural Networks (DRCNNs), Transfer Learning, Explainable Artificial Intelligence (XAI), Mammography
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	PREDICTION OF POLYCYSTIC OVARIAN DISEASE IN MEDICAL DATA USING DEEP LEARNING MODELS
Author:	R. PARVATHI, DR.P.GEETHA
Abstract:	Objectives/Backgrounds: In modern times, polycystic ovarian disease, or PCOD, is a major factor in women's lives. The primary causes of PCOD are a mix of genetic predispositions and hormone imbalance. Every month, the two ovaries in a typical menstrual cycle will release mature, fertilized eggs in turn. Methods/Statistical Analysis: For the prepossessing, the PCOD dataset is obtained in a.csv file format from the Kaggle repository. Pre-processing involves removing irrelevant data, adding missing values, and so on. The prediction then receives the final product as its input. Findings: Basic characteristics like age, height, and weight are considered for the prediction, along with specific features like I beta and II beta HCG, FSH, LH, endometrial thickness, and screening to see if the patient is pregnant. To determine the precision of the algorithms, the data set that has been processed is categorized using Deep Learning Models like DNN, RNN, and CNN. Classification metrics including precision, recall, and f-measure values are used to compare the performance of the three techniques; DNN performs better than the other two. Improvement: Subsequently, further categorization techniques were employed to locate vast amounts of data.
Keywords:	Polycystic Ovarian Syndrome, DNN, RNN, CNN, Polycystic Ovarian Disease Prediction., Precision, f-measure, recall, Accuracy.
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	ADAPTIVE ROUTING PROTOCOL FOR BUOYANT WIRELESS SENSOR NETWORK (B-WSN) USING HAMSTER OPTIMIZATION-BASED MAXPROP (HOMP)
Author:	DIVYA JOSE J, Dr. D. VIMAL KUMAR
Abstract:	This study presents adaptive routing protocol proposed for Buoyant Wireless Sensor Networks (B-WSNs) through the integration of Hamster Optimization and Max Prop (HOMP). Buoyant WSNs operate in challenging aquatic environments, posing unique challenges such as underwater communication, limited energy resources, and prolonged network operation. To address these challenges HOMP the routing protocol is proposed which optimizes routing decisions and resource utilization in buoyant WSNs. Through extensive experimentation, we evaluate HOMP's performance in terms of packet delivery ratio, throughput, energy consumption, and network lifetime, comparing it with existing protocols. Results demonstrate that HOMP consistently outperforms other protocols, offering superior efficiency and reliability in data transmission. These findings have significant implications for applications in oceanography, environmental monitoring, and marine exploration, where reliable underwater sensing and monitoring are essential. By leveraging the capabilities of HOMP, buoyant WSNs can achieve greater accuracy, coverage, and longevity, enabling more effective data collection and analysis in aquatic environments.
Keywords:	Wireless Sensor Networks, Buoyant WSN, Routing Protocols, Hamster Optimization, MaxProp, Underwater Communication, Energy Efficiency, Network Lifetime.
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	AN INTEGRATED ANOMALY DETECTION FRAMEWORK FOR STREAM DATA USING EXTENDED LOF WITH WINDOW METHOD
Author:	AKURI SREE RAMA CHANDRA MURTHY , Dr. CH. VENKATA NARAYANA
Abstract:	Multivariate time series and data streams are closely linked, but the latter typically show a larger time dependence and do not require real-time processing. Numerous fields, including network intrusion detection, financial fraud detection, and defect detection in industrial and infrastructure systems, depend on the ability to identify anomalies in streaming data. The majority of anomaly detection (AD) algorithms now in use work well with static data, when all accessible information is available at the time of detection. However, they are unable to handle dynamic data streams. Our study's EM-W-LOF (Extended Local Outlier Factor) algorithm, which depends on Expectation Maximization and the window model, outperforms traditional techniques and offers an effective way to detect anomalies in data streams. Expectation maximization (EM) is a method applied to process data rectification. To lower the false alarm rate, data windows are incorporated as update units. Several tests are conducted here to distinguish between candidate and actual anomalies. The enhanced EM-W-LOF's false positive rate demonstrates its benefit. Additionally, data points of detected actual anomalies are immediately deleted by the suggested technique. Through practical studies with both synthetic and actual data sets, we examined the enhanced algorithm's performance as well as the sensitivity of specific parameters. The experimental findings show that, in comparison to the standard algorithms and their enhancements, the suggested improved algorithm performed better in terms of both detection rate and false alarm rate
Keywords:	Window Model, Data Streams, Anomaly Detection, Incremental Local Outlier Factor Algorithm
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	SEER: SECURED ENERGY EFFICIENT ROUTING ALGORITHMS FOR ATTACKS IN WIRELESS SENSOR NETWORKS
Author:	DR.N.GAYATRI, D. CHAITHANYA, CH V RAGHAVENDRAN, DR. PARVATHI MALEPATI, K. SHYAM SUNDER REDDY, DR. M. KIRAN KUMAR, DR. P.NARESH
Abstract:	The main focus in WSN (Wireless Sensor Networks) is to perform an energy efficient routing for path finding from source to destination in communication channel. The existing studies performed well, but suffering from security aspects in terms of various attacks on networks. In this paper we focused more on how to handle various network attacks along with proper network routing. WSN deals with sensors hence involvement of energy consumption factors to be updated and can also be depends on topologies using for WSN. Node behavior is the major challenge which causes congestion in communication. We proposed secured energy efficient routing (SEER) algorithm to brief the mentioned problems. This method constructs an adaptive trust based secured model for network attacks such as black hole, sinkhole, hello flood and selective forwarding. The SEER algorithm uses various trust values to probe the attacks. Penalty mechanism can be used for identifying bad nodes in and around routing. By using multi hop routes predicted wormhole problems in network communication. Finally, the comparative study shows that our results are better than the others in saving of energy, detecting bad node behavior and prohibiting other network attacks.
Keywords:	WSN, Network Attacks, Secure Routing, Iot, Trust Based Model, Penalty Mechanism, Cluster Head.
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	UTILIZING EXPLORATORY DATA ANALYSIS AND MACHINE LEARNING TO ENHANCE LEARNING QUALITY IN THERMODYNAMICS
Author:	ARWIZET KARUDIN, DESMARITA LENI, REMON LAPISA, YUDA PERDANA KUSUMA
Abstract:	Students often face difficulties in understanding abstract concepts in thermodynamics, such as thermal system efficiency, temperature distribution, and inter-variable interactions in thermal phenomena. These challenges are compounded by traditional teaching methods, which typically rely only on text and mathematical calculations. This research aims to develop a lab data-based visualization model to enhance students' understanding of thermodynamics concepts. The model utilizes experimental data collected by Mechanical Engineering students at Universitas Negeri Padang on various instructional apparatus, including steam power plants, internal combustion engines, wind turbines, and crank mechanisms, gathered between 2022 and 2024. This visualization model includes several key features: a CRUD-based data storage system using Pymongo, 2D data visualization with Pygwalker, 3D visualizations in contour and surface diagrams, correlation analysis using heatmaps, and machine learning-based predictions with PyCaret. Evaluation results indicate that this model significantly improves students' comprehension of abstract thermodynamics concepts. Based on assessments from four thermodynamics expert lecturers, the information accuracy aspect received an average score of 4 (Good), visualization quality received an average score of 4.25 (Very Good), and ease of use received the highest average score of 4.5 (Very Good). The relevance of the model to learning objectives received an average score of 3.75, slightly below the Good category, indicating room for improvement in terms of educational relevance. Overall, this research demonstrates that integrating EDA and machine learning through the developed visualization model is effective in supporting more interactive and data-driven thermodynamics learning, aligned with the needs of education in the digital era.
Keywords:	Exploratory Data Analysis, thermodynamics, machine learning, correlation, analysis
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	A NOVEL LSB8 METHOD FOR MESSAGE STEGANOGRAPHY USINGDIGITAL SPEECH SIGNAL
Author:	HISHAM ALRAWASHDEH , AYMAN AL-RAWASHDEH , MAHMOUD QASAYMEH
Abstract:	Providing a simple and secure way to preserve confidential messages is crucial, given the widespread exchange of messages across various communication channels. This research paper presents a straightforward and effective method for protecting secret messages. The approach utilizes a speech file as the cover medium, enhancing the hiding capacity by enabling the concealment of a message that matches the size of the cover media. The method hides one character from the secret message using every sample of the speech file. The sample values of the speech file will be converted into 64-bit binary numbers, with the 8 least significant bits of each sample reserved for embedding the 8-bit character. Replacing the eight least significant bits of each sample with the binary value of the character will minimally impact the sample's value. As a result, the stego sample value will closely resemble the original cover sample value, ensuring excellent quality in the stego file. The mean square error will be close to zero. The suggested method will be tested using several messages and cover speech files. The resulting experimental data will be analyzed to demonstrate the method's enhancements in increasing hiding capacity, speeding up the message embedding process, and minimizing the error between the cover and stego speech, effectively making the stego speech indistinguishable from the original.
Keywords:	Steganography, Covering speech, Stego Speech, LSB, LSB2, LSB8, MBM, CBBM, Quality.
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	TOWARDS ROBUST TAI LANGUAGE ASPECT-BASED SENTIMENT ANALYSIS: A WEAK-SUPERVISION APPROACH
Author:	KANKANA DUTTA , RIZWAN REHMAN
Abstract:	Speech signals carry extensive information about the speaker, including various non-linguistic elements such as sentiment, emotion, and intent. Analyzing these aspects has garnered significant attention due to its wide-ranging applications in fields such as human-computer interaction, mental health assessment, and social media analytics. This paper presents a novel approach to sentiment analysis for the Tai-Phake language, spoken by the Tai-Phake community of Assam, a language with limited linguistic resources. We propose a Convolutional Neural Network (CNN) model that classifies sentiments by leveraging Mel-Frequency Cepstral Coefficients (MFCC) features extracted from speech data. The scarcity of linguistic resources for Tai-Phake posed substantial challenges, necessitating innovative solutions to develop an effective sentiment analysis tool. Our study also benchmarks the performance of the proposed CNN model against two other popular methods: a Multi-Layer Perceptron (MLP) classifier and a Long Short-Term Memory (LSTM) network. Extensive experiments reveal that the CNN model achieves superior performance, with higher accuracy and robustness in classifying sentiments compared to the MLP and LSTM classifiers. This indicates the effectiveness of convolutional architectures in capturing the intricate patterns in speech signals relevant to sentiment analysis. The findings underscore the potential of CNNs in handling sentiment classification tasks in resource-constrained languages like Tai-Phake, paving the way for further advancements in the field of speech-based sentiment analysis.
Keywords:	Machine Learning methods, Neural Networks, CNN, Sentiment Analysis, MFCC
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	OPTIMIZED DEEP AUTO-ENCODER INTEGRATED WITH QUADRATIC SUPPORT VECTOR MACHINE FOR ENHANCED CREDIT CARD FRAUD DETECTION
Author:	DEEPIKA SIRMORIA, S. SENTHIL
Abstract:	Credit card fraud detection is a challenging research area in which many factors influence the performance of methods. The majority of credit card fraud detection systems relied on an examination of previous transactions. As long as changes in customer behaviour, we need different fraud detecting strategies. Every year, millions of rupees are lost due to a lack of awareness of changes and the fraud detection. For minimizing such loss, we need to develop and implement efficient framework which can adapt non-linear behaviours of transactions. In this paper, we used efficient and optimal deep auto encoders (DA) for optimal feature selection and then these features are given to nonlinear learning approach i.e. quadratic support vector machine (QSVM) for classifying the transaction as fraudulent or not. In this approach, iterative fine-tuning process is considered in testing phase which can update parameters of training model. The proposed method is tested using various training dataset ratios and the calculated sensitivity, specificity, and accuracy measurement parameters. We use real world dataset for classifying fraudulent and non-fraudulent transaction by focusing on the low-dimensionality, optimal feature selection and fine tuning. The proposed DA-QSVM solution achieves comparable performance values with existing state-of-the-art and costly solutions.
Keywords:	Credit Card Fraud, Non-Linearity, Fine- Tuning, Auto Encoder, Quadratic SVM.
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	A NEW APPROACH TO NEEDS ANALYSIS: POWER PLANT MODELLING AND ANALYSIS IN SUSTAINABLE RENEWABLE ENERGY LEARNING
Author:	RAHMANIAR , AGUS JUNAIDI , RIAN FARTA WIJAYA
Abstract:	This research examines renewable energy modelling and analysis to develop future power plants in higher education. A new approach in needs analysis through modelling and analysis of renewable energy generation to design future power plants. The case study focuses on a solar power plant and a hydroelectric power plant in Tomuan Holbung Village Bandar Pasir Mandoge Asahan Regency, North Sumatra, Indonesia. The potential energy sources of hydropower and solar power plants in Tomuan Holbung Village were mapped and measured and then simulated, resulting in a hydropower potential of 5.92 MW. As for the potential of solar energy from the measurement results and obtained average daily radiation of solar energy reaches 5.9 kWh/m2. This research also explores the application of the research results in a real problem-based e-module for learning. The practicality of the real problem-assisted e-module model has been tested in a small class test on 4 student respondents who participated in the study, showing the results of the practicality of using the renewable energy e-module of 80% declared practical. The concept of mapping, modelling and analysis of real situations applied to learning becomes a reference in understanding future electrical energy needs whose potential is spread across several regions in Indonesia. This approach offers new insights and methodologies that can be adapted for sustainable power plant design and enriches the learning experience in the field of renewable energy.
Keywords:	Five Renewable Energy; Hydro Power;Photovoltaic;Sustainable Power Plant; Riil Problem Based Learning
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	COMPETENCY AND GUIDELINE FOR ASSESSING READINESS TO BE AN INFORMATION TECHNOLOGY AUDITOR IN THAILAND
Author:	KASIPAT THANITTHANAKHUN , WATCHAREEWAN JITSAKUL
Abstract:	In the digital era, the role of the IT auditor has become even more critical due to the increased reliance on digital technologies, the proliferation of data, the growing complexity of IT systems, and the growth of computer crime. To be a qualified IT auditor, the IT auditor must have good competencies that will result in quality audit work and results as determined by the organization. IT audit competency influences audit quality. Currently, there is a shortage of IT auditors, with many interested people needing more guidelines to become IT auditors. Therefore, this research aimed to develop a list of essential IT Audit competencies and guidelines for assessing such competencies in Thailand to help produce a new generation of qualified IT auditors. The research was divided into 3 parts: 1) Competency synthesis by in-depth interviews with 10 experts and sending online questionnaires to auditing professionals, with 400 responses. Statistics were used for analysis, including mean, standard deviation, and exploratory factor analysis, using KMO (Kaiser-Meyer-Olkin) statistics and Bartlett's Test of Sphericity 2) Development of competence indicators by holding focus group meetings with 10 experts. 3) Development of an IT auditor competency assessment system, which was evaluated for suitability and satisfaction by 10 experts and 30 users. The evaluation results were excellent.
Keywords:	IT audit, IT auditor, IS Audit, IS Auditor, IT Audit competence, IT Audit Competency Assessment System.
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	ADVANCED DISEASE DETECTION USING HYBRID CNN WITH LSTM AND GRU MODELS: A DEEP LEARNING APPROACH
Author:	NAYAN JADHAV, Dr. AZIZ MAKANDAR
Abstract:	The area of medicine has seen a tremendous transformation as a result of the incorporation of cutting-edge technologies like machine learning. This study combines a hybrid structure combining Convolutional Neural Networks (CNN) with Long Short-Term Memory (LSTM) networks and Gated Recurrent Unit (GRU), to enhance the accuracy of sickness analysis and remedy prediction. The aim is to leverage CNN for feature extraction from imaging information and LSTM for temporal pattern reputation in time-series patient data and GRU for sequential data. Medical datasets for illnesses which include lung cancer, brain tumors, pneumonia, tuberculosis, skin cancer and breast cancers had been used, with a pre-processing that protectes records balancing, normalization, and augmentation. The CNN-LSTM version validated superior overall performance as compared to CNN-GRU and traditional models, with full-size enhancements in accuracy throughout more than one illnesses. Key findings indicate that the CNN-LSTM model achieves an accuracy of an average 99.2% and 98.6% for CNN-GRU, highlighting its efficacy in complicated diagnostic eventualities. This research underscores the ability of mixing CNN with LSTM and GRU for advanced disease diagnosis, presenting a scalable and powerful method for healthcare packages. The study's effects advocate better avenues for future findings, together with actual-time medical diagnostics and the enlargement of the dataset to include extra numerous clinical conditions.
Keywords:	Healthcare, Convolutional Neural Network, Disease Diagnosis, Long Short-Term Memory, Gated Recurrent Unit
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	AN APPROACH FOR VULNERABILITY DETECTION IN WEB APPLICATIONS USING GRAPH NEURAL NETWORKS AND TRANSFORMERS
Author:	MOHAMMED YAHAYA TANKO, ABU BAKAR MD SULTAN, MOHD HAFEEZ OSMAN, HAZURA ZULZALIL
Abstract:	The increasing complexity of software systems and rising security concerns due to open-source package vulnerabilities have made software vulnerability detection a critical priority. Traditional vulnerability detection methods, including static, dynamic, and hybrid approaches, often struggle with high false-positive rates and limited efficiency. Recently, graph-based neural networks (GNNs) and Transformer models have shown potential in improving vulnerability detection accuracy by representing code as graphs that capture syntax and semantics. This paper introduces a hybrid framework combining a Gated Graph Neural Network (GGNN) and Transformer encoder to leverage multiple graph representations: Abstract Syntax Tree (AST), Data Flow Graph (DFG), Control Flow Graph (CFG), and Code Property Graph (CPG). The GGNN extracts graph-level features, while the Transformer enhances sequential context understanding within the graph-encoded data. The model uses these capabilities to detect vulnerabilities in function-level code snippets. Evaluation of our framework on the OWASP WebGoat dataset demonstrates the effectiveness of different graph representations across five major vulnerability types: command injection, weak cryptography, path traversal, SQL injection, and cross-site scripting. Experimental results show that the GGNN+CPG configuration consistently yields high recall for cryptographic weaknesses, while GGNN+CFG excels in detecting control-based vulnerabilities, such as command injections. The integration of GGNN and Transformer models leads to notable enhancements in accuracy, precision, recall, and F1-score across all vulnerability types, with each graph representation contributing unique insights into code structures and vulnerability patterns. These findings highlight the potential of hybrid GNN-Transformer frameworks in enhancing code vulnerability detection for cybersecurity applications.
Keywords:	Vulnerability Detection; Graph Neural Networks; Software Security, Web Application Security, Transformers In Cybersecurity
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	BEYOND THE NOISE: UNVEILING MEANINGFUL PATTERNS IN MOBILE PAYMENT DATA USING FUZZY LOGIC AND SVM
Author:	GAITH M. ABASHES , BASIMA ELSHQEIRAT , AHMAD A. ABU-SHAREHA , MOHAMMAD ALSHRAIDEH
Abstract:	Big Data refers to vast and complex datasets that require processing and analysis to uncover valuable insights beneficial to businesses and organizations. This has become a fundamental approach for acquiring, processing, and analyzing large volumes of data to extract useful information. Big Data encounters challenges such as Volume, Velocity, Variety, Variability, and Value. Preprocessing and analysis are crucial for obtaining quality information that supports accurate decision-making. Many organizations now handle extensive data, often called "big data," due to its size, speed, and diverse formats, revolutionizing decision support and data management. The primary challenge of big data is to extract values for decision-making, prediction, and service improvement. The proposed framework focuses on preserving financial data quality by clustering with fuzzy Logic and Support Vector Machines (SVM). It comprises four layers: data collection in the first layer, preprocessing data in the second layer, which involves cleaning and mapping semi-structured data to establish relationships; the third layer applies a fuzzy controller and classification to generate rules; the fourth and final layers; and data reduction and classification employing SVM clustering to create distinct clusters for meaningful and predictive outcomes. This study addressed two key challenges: 1. Extracting meaningful information from big data: This study aimed to extract valuable insights from the vast data generated in the mobile payment sector. 2. Utilizing big data techniques with SVM clustering and leveraging fuzzy Logic, this study employs Support Vector Machine (SVM) clustering to identify patterns and relationships within the data. Additionally, fuzzy Logic was incorporated to extract the rules and connections among the attributes. We experimentally validate the proposed approach using financial company data from the mobile payment industry.
Keywords:	Big Data, Decision Support, Prediction, Fuzzy Logic, Support Vector Machines (SVM).
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	A HYBRID MODEL COMBINING GRAPH NEURAL NETWORKS, REINFORCEMENT LEARNING, AND AUTOENCODERS FOR AUTOMATED CODE REFACTORING AND OPTIMIZATION
Author:	RAGHUPATHY DURGA PRASAD, Dr. MUKTEVI SRIVENKATESH
Abstract:	This research develops a cutting-edge hybrid deep learning architecture, blending Graph Neural Networks (GNNs), Reinforcement Learning (RL), and Autoencoders, to optimize and refactor code automatically. GNNs are employed to capture hierarchical and structural relationships within code, RL iteratively optimizes the refactoring process based on performance metrics, and Autoencoders compress code representations to reduce redundancy and enhance efficiency. The proposed model outperforms standalone GNN, RL, and Autoencoder models as well as traditional heuristic-based methods, achieving an accuracy of 92.5%, precision of 91.8%, recall of 90.7%, and F1-score of 91.2%. Experimental results also demonstrate substantial improvements in code complexity metrics, including a 35.2% reduction in cyclomatic complexity, 28.7% fewer lines of code, and a 40.3% decrease in code coupling, enhancing readability and maintainability. Furthermore, the model operates with an average runtime of 1.5 seconds and memory usage of 150 MB, significantly outperforming baseline approaches. These findings affirm the model’s efficacy in delivering high-quality, resource-efficient code refactoring solutions, making it a robust tool for modern software engineering practices.
Keywords:	Hybrid Deep Learning, Code Refactoring, Graph Neural Networks (GNNs) Reinforcement Learning (RL), Autoencoders, Software Optimization.
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	SENTIMENT ANALYSIS OF THE STATE GLOBAL ISLAMIC ECONOMY ON TWITTER WITH SUPPORT VECTOR MACHINE RAPID MINER
Author:	WAHYU SARDJONO , ALDO LOVELY ARIEF SUYOSO , ERMA LUSIA
Abstract:	This study uses the Support Vector Machine (SVM) algorithm on the RapidMiner platform to collect sentiment towards the State of the Global Islamic Economy (SGIE) in Twitter data related to the 2024 Vice Presidential candidate debate. Initial data obtained through Twitter crawling reached 22,673 entries. The data used for analysis was 12,135 entries after duplicate data was removed, then about a quarter of the data, or 2,925 entries were labeled as training data. The dataset was taken from Twitter with the search keyword SGIE in December 2023. SVM analysis predicts the negative sentiment class with an accuracy level above 90%. The dominance of negative sentiment and evaluation shows quite good accuracy of 76.30%, precision of 67.92%, and recall of 99.69%. These results provide important insights into public responses to global Islamic economic issues, highlighting the relevance of sentiment analysis methods in understanding the dynamics of public opinion and their contribution to developing future policies and strategies.
Keywords:	State of the Global Islamic Economy (SGIE), 2024 Vice Presidential Candidate Debate, Sentiment Analysis, SVM, Twitter, RapidMiner.
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	A HYBRID QUANTUM-INSPIRED CNN ARCHITECTURE FOR EFFICIENT AND ACCURATE BRAIN TUMOR CLASSIFICATION WITH EXPLAINABILITY ANALYSIS
Author:	SHEETHAL M. S , DR. P. AMUDHA
Abstract:	Brain tumors are considered one of the serious types of cancer and catching them early is crucial for better patient results. The use of Magnetic Resonance Imaging (MRI) plays a role in spotting and categorizing brain tumors; however, it can be tough and time consuming for radiologists to interpret these images. To address this issue, this study introduces a Quantum Inspired Convolutional Neural Network (QCNN) that integrates concepts from quantum mechanics, such as superposition and entanglement, with traditional Convolutional Neural Network (CNN) architectures. This innovative approach enhances feature extraction capabilities, allowing the model to recognize intricate patterns within MRI images more effectively. The QCNN achieved impressive results, reaching a peak validation accuracy of 99.44% and demonstrating significantly improved classification for various types of brain tumors, including Glioma and Meningitis. These findings highlight the potential of the QCNN framework to revolutionize tumor detection, offering radiologists a powerful tool that enhances diagnostic accuracy and efficiency in clinical practice, ultimately leading to better patient care and outcomes in the medical imaging field.
Keywords:	Quantum, CNN, Architecture, Brain Tumor, Classification
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	PREDICTIVE ANALYSIS ON DEMONETIZATION DATA USING SUPPORT VECTOR MACHINE TECHNIQUE
Author:	KALIAPPAN M, MARIAPPAN E, RAMNATH M, KARPAGAVALLI C, ANGEL HEPZIBAH R, VIMAL S
Abstract:	The Unexpected announcement of demonetization of 500 and 1000-rupee note caused chaos in the cash-dependent economy which resulted in a lot of small-scale industries and entrepreneurs who lost their livelihood due to these events. However, the main intent of this Scheme is to eradicate black-market money, to combat inflation and move towards the betterment of our society. There have been a lot of mixed views towards this issue. In this paper, we develop predictive Analysis on Demonetization Data using SVM approach (PAD-SVM). The proposed system involved three stages including preprocessing stage, descriptive analysis stage, and descriptive analysis. The Preprocessing stage involves cleaning the obtained data, performing missing value treatment and splitting the necessary data from the tweets. The Descriptive analysis stage involves finding the most influential people regarding this subject and performing analytical functionalities. These analytical functionalities include striping the already processed tweets, cleansing them from special characters, lemmatizing the tweets. Semantic analysis is performed to find the sentiment values of the users and to find the compound polarity of each tweet. Once the polarity scores are calculated, categorize these tweets as “POSITIVE”, “NEGATIVE” and “NEUTRAL”. Descriptive analysis is performed to view the current mindset of people and the society reacts to the issue in the current time. Predictive analysis is performed to predict the mindset of people which may change. This analysis is performed to find out the overall viewpoint of the society and their view may change in the near-future in regarding to the scheme of demonetization as well.
Keywords:	Descriptive Analysis, Predictive Analysis, Support Vector Machine, Sentiment Analysis
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	PREDICTION OF STOCK PRICE USING HYBRID NEURAL NETWORK: A CASE OF COAL PRODUCTION COMPANY
Author:	TOSSAPOL KIATCHAROENPOL , SAKON KLONGBOONJIT
Abstract:	Stock market prediction is a critical issue in the field of economics. As machine learning technologies advance, an increasing number of algorithms are being utilized to forecast stock price movements. Nonetheless, predicting stock market trends remains a challenging task due to the inherent noise and volatility in stock market data. This paper addresses this challenge by proposing a novel hybrid neural network model designed to predict stock market prices using parameters related to commodity prices and stock indices. A case study company is mainly in coal production business in Thailand, which produce coal, sale, distribute and operate coal-fired power plants as well. The Multiple Linear Regression (MLR) and Back propagation neural network (BPNN) as traditional prediction technique are employed to comparatively investigate the accuracy and performance of the proposed HNN. Experiment results show that the prediction accuracy of HNN is superior to MLR but similar to that of the BPNN model. However, HNN has a good performance both in accuracy, speed and practice. It can help investing analysts and investors make their wise decisions.
Keywords:	Artificial Neural Network, Stock Price Prediction, Coal production, Hybrid Neural Network.
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	ADVANCED DEEP LEARNING TECHNIQUES FOR DIABETIC RETINOPATHY DETECTION USING CLAHE-GAMMA-UNSHARP HYBRID ENHANCEMENT
Author:	ARPITA NIBEDITA , PRABHAT KUMAR SAHU , SRIKANTA PATNAIK , KUMAR SURJEET CHAUDHURY
Abstract:	Diabetic Retinopathy (DR) is a leading cause of blindness worldwide, affecting individuals of all ages. Each year, a significant number of people suffer vision loss due to DR. The diagnosis of DR requires precision, as even minor errors can lead to severe consequences. Misdiagnosis is common and can significantly increase patient morbidity. To address these challenges, this paper presents a comprehensive study on the classification of Diabetic Retinopathy (DR) images using three deep learning models: ResNet101V2, InceptionResNetV2, and a custom CNN model. The goal is to evaluate the performance of these models and improve generalization through a stacking ensemble approach with a logistic regression meta-learner. Each model was fine-tuned to enhance feature extraction and classification performance. ResNet101V2 achieved a peak training accuracy of 95.95% but exhibited overfitting, with a test accuracy of 79.81%. InceptionResNetV2 achieved an exceptional training accuracy of 99.91%, though its test accuracy was slightly better at 81.31%. The custom CNN, a simpler architecture, demonstrated a balanced performance with 87.24% training accuracy and 86.63% test accuracy, highlighting its strong generalization capabilities. A stacking ensemble was implemented to improve prediction accuracy, combining the outputs of the three models using pseudo-labeling techniques. The meta-learner enhanced the overall classification accuracy, leveraging the strengths of the individual models. This ensemble approach proved effective in improving model generalization to unseen data. Future work will focus on mitigating overfitting in more complex models, exploring transformer-based architectures, and applying these models to real-world clinical datasets for DR detection.
Keywords:	Diabetic Ratinopathy, ResNet101V2, InceptionResNetV2, Custom CNN, Stacking Ensemble, Logistic Regression, Random Forest, Gradient Boosting, Decision Tree
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	TROPICAL CYCLONE DETECTION IN COASTAL AREA USING RECURRENT ALL-PAIRS FIELD TRANSFORMER FEATURE EXTRACTION AND 3D-CNN CLASSIFICATION MODEL FROM GEOSTATIONARY SATELLITE DATA
Author:	NAGU MALOTHU , DR.V.V.K.D.V.PRASAD, DR. B.T.KRISHNA
Abstract:	High amplitude depending on the wind's top speed and the categorization of intensity are both used in cyclone classification and prediction models. The whole spectrum of ideal features needed for classification are depreciated by the computational limits combined with the production of those intensities, cyclone categorization, and forecast, making accurate representation less likely. There is a bias that varies depending on the Tropical Cyclone (TC) center and shape because there is no standardized way for calculating TC intensity and the most popular method uses a manual computation employing satellite-based weather imagery. The Lucas-Kanade optimal flow based Recurrent All-Pairs Field Transformer (LK RAFT) is thus the main focus of this work for feature extraction. It collects properties at the pixel level, creates 4D comparative quantities at multiple scales for all possible pixel resolutions, and retrieves these values through iteratively notifying a flow field via a recurrent unit. Furthermore, three-dimensional convolutional neural networks (3D-CNN) were used to investigate the correlation between TC intensity and multi-spectral geostationary satellite images. Additionally, we used CNN visualization tool, to examine the properties of multi-spectral satellite-based TC pictures according to intensity. Using images from cyclone samples like OCKHI DEC2017 and VARDAH DEC2016, empirical assessment of the proposed LK RAFT 3DCNN method is carried out with the consideration of variables like accuracy, precision, recall, mean square error and F1-score.
Keywords:	Tropical Cyclone (TC), Prediction, feature extraction, satellite images, optimal flow
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text

Title:	MACHINE LEARNING APPLIED TO THE EARLY DETECTION OF HEART ATTACKS IN PATIENTS WITH TERMINAL CHRONIC KIDNEY DISEASE
Author:	BRYAN WALTER MEJIA MANZANARES, DIEGO RICHARD RIVERA DEMANUEL, YASIEL PÉREZ VERA
Abstract:	Currently, worldwide one in ten adults suffers from chronic kidney disease. The presence of renal lesions shows this disease. Causes the death of at least 2.4 million people a year; it is therefore appropriate to study and develop solutions to reduce the possibility of death. This study aims to develop a predictive model to aid in detecting heart attacks in patients suffering from this disease. To achieve this, six algorithms, Random Forest, XgBoost, Adaboost, Decision Tree, Support Vector Machine, and Gradient Boosting, were applied to construct the model. Statistical comparison was then performed using F1-score, Accuracy, Precision, Area Under the Curve, Recall, MCC and Kappa metrics to detect the best model. Adaboost was obtained as the best algorithm for the construction of models of the same nature. As a result, a model was developed to help predict a heart attack in people with chronic renal failure. This model allows classifications or predictions of this forecast to be made with good results and helps to reduce the risk of death in patients due to its high percentage of effectiveness. It could also be a starting point for future models that treat the same disease.
Keywords:	Machine Learning, Predictive Models, Adaboost, Random Forest, SVM, Decision Tree, Chronic Kidney Disease, Heart Attack.
Source:	Journal of Theoretical and Applied Information Technology 15^th January 2025 -- Vol. 103. No. 1-- 2025
Full Text