International Journal of Computer Communication and Informatics

Credit Card Fraud Detection Using State-of-the-Art Machine Learning and Deep Learning Algorithms

2025-07-29T06:11:16+00:00

Credit card fraud is a major concern for both financial institutions and consumers, leading to significant financial losses and a decline in trust. With the rise in online transactions and increasingly sophisticated fraudulent schemes, there is a pressing need for strong and effective fraud detection systems. This research explores how machine learning and deep learning algorithms, particularly Random Forest (RF) and K-Nearest Neighbors (KNN), can be applied to detect credit card fraud. The main goal is to assess and compare how well these algorithms perform in accurately spotting fraudulent transactions while keeping false positives to a minimum. To carry out this research, we use a publicly available dataset of credit card transactions, which is marked by an imbalanced class distribution, where fraudulent transactions are far fewer than legitimate ones. We apply various preprocessing techniques, such as data cleaning, feature scaling, and addressing class imbalance through resampling methods like SMOTE (Synthetic Minority Over-sampling Technique), to improve data quality and model performance. Random Forest is a powerful ensemble learning method that uses a collection of decision trees to boost prediction accuracy and cut down on overfitting. K-Nearest Neighbors (KNN) is a straightforward, instance-based learning algorithm that classifies transactions by looking at the majority class of their k-nearest neighbours in the feature space. To evaluate how well both algorithms perform, we look at various metrics like precision, recall, F1-score, and the Area Under the Receiver Operating Characteristic Curve (AUC-ROC). The findings show that Random Forest typically outshines K-Nearest Neighbors in overall accuracy and F1-score, especially when dealing with imbalanced datasets. This research emphasizes the need to tackle class imbalance and choose the right evaluation metrics for effective fraud detection.

Comparative Analysis on Different Deepfake Detection Techniques

2025-07-29T06:11:16+00:00

Advancements in deep learning have led to the emergence of highly realistic AI-generated videos known as deepfakes. These videos utilize generative models to expertly modify facial features, creating convincingly altered identities or expressions. Despite their complexity, deepfakes pose significant threats by potentially misleading or manipulating individuals, which can undermine trust and have repercussions on legal, political, and social frameworks. To address these challenges, researchers are actively developing strategies to detect deepfake content, essential for safeguarding privacy and combating the spread of manipulated media. This article explores current methods for generating deepfake images and videos, with a focus on facial features and expression alterations. It also provides an overview of publicly available deepfake datasets, crucial for developing and evaluating detection systems. Additionally, the research examines the challenges associated with identifying deepfake face swaps and expression changes, while proposing future research directions to overcome these hurdles. By offering guidance to researchers, the document aims to foster the development of robust solutions for deepfake detection, contributing to the preservation of the integrity and reliability of visual media.

Harnessing Artificial Intelligence for Disease Detection and Rapid Drug Discovery: A Path to Accelerated Medical Responses

2025-07-29T06:11:16+00:00

The history of Artificial Intelligence (AI) in drug discovery spans decades, from rule-based systems to sophisticated machine learning and deep learning algorithms. Early applications included virtual screening and QSAR modeling, which paved the way for data-driven drug development. Today, systems like IBM Watson Health and DeepMind's AlphaFold are good at analyzing medical data, predicting molecular interactions, and accelerating the design of novel drugs. Yet in most AI solutions that already exist, they usually only solve the specific tasks rather than formulating a comprehensive framework in emerging disease management. This paper proposes the integration of disease symptom data, pathogen-level analysis, and treatment prediction via an AI-driven model about diseases with symptoms such as cold, cough, or fever. The system correlates new pathogens with stored datasets and identifies potential medicine combinations for rapid testing and refinement, thereby significantly reducing the timelines for drug development. Hence, this approach addresses the severe need for scalable, fast-response solutions in managing infectious diseases and future pandemics.

A Comparative Analysis of Machine Learning Models for Stroke Prediction

2025-07-29T06:11:16+00:00

Stroke is a leading global health burden, and there is an urgent need for improvement in risk prediction and treatment. This paper examines the capability of several machine learning algorithms, including Decision Trees, Random Forests, Neural Networks, Support Vector Machines (SVMs), Elastic Nets, and Lasso, to predict stroke risk on four cardiovascular and stroke datasets. The results indicate that Decision Trees and Random Forests are always better than Neural Networks, although Neural Networks show promising accuracy. SVMs are consistent, while the Elastic Net and Lasso models give average results.

A Low Power Hybrid DCO Using Three Transistor (3-T) XNOR Gate, CMOS and Pseudo-NMOS Inverter

2025-07-29T06:11:16+00:00

This research article presents a comprehensive investigation of three-bit hybrid-digitally controlled ring oscillator (HDCRO) implemented with TMSC 90nm CMOS technology. The hybrid circuit HDCRO comprises of three distinct delay stages, namely XNOR-based inverter, a CMOS inverter, and a Pseudo-NMOS inverter, all of which have been designed utilizing an inversion MOS varactor (IMOS). Furthermore, the investigation explores the output frequency variation in the load element of the HDCRO by adjusting the capacitance of the digitally controlled MOS varactors. This frequency variation occurs as a result of changing the digital control bits of the MOS varactors at a supply voltage of 0.7 V. The proposed HDCRO demonstrates an oscillation frequency range of 2.558 GHz to 2.649 GHz, with power consumption varying from 3.638 mW to 1.046 mW, and phase noise from -68.070 dB@1 MHz to -67.654 dB@1 MHz relative to the central oscillation frequency. Moreover, by applying a supply voltage variation between 0.5 V and 1 V, a wider frequency tuning range of 1.238 GHz to 4.438 GHz is achieved. This extended tuning range exhibits power consumption variation from 2.785 µW to 54.66 mW, and phase noise from -68.812 dB@1 MHz to -65.445 dB@1 MHz relative to the central oscillation frequency. In summary, this study presents a novel HDCRO architecture that demonstrates excellent performance in terms of frequency range, power consumption and phase noise. The proposed design offers advantages of high speed, low-power and good frequency range; thus has a promising prospect of application in high-performance integrated circuits.

Semantic Tag Clustering to Alleviate the Cold Start Problem in Learning Resource Recommendation: A Case Study on Delicious Dataset

2025-07-29T06:11:16+00:00

This study explores a methodology for recommending learning resources, demonstrated through a case study on the Delicious dataset. Tags, representing keywords assigned to describe content, are semantically clustered using K-Means. Sentence Transformers are employed to generate dense vector representation of these tags, enabling more effective clustering. The system identifies meaningful tag groups to deliver relevant recommendations, even in the absence of user interaction history, effectively addressing the cold-start problem through predefined tag profiles. The proposed methodology personalizes resource recommendations for Deaf and Hard of Hearing (DHH) learners by leveraging their profile and resource Meta data. It enhances resource search during the cold start phase by identifying the most relevant tag cluster that matches with the learner’s search query and retrieving preferred content based on the learner profile. Future extensions could incorporate dynamic preferences that evolve over time, enabling more adaptable and personalized recommendations. This work provides a robust foundation for clustering the resources based on their semantic meaning, thereby improving content-based search and retrieval of relevant learning resources.

Automated System to Preventing Social Security Fund Misuse by Identifying Deceased Beneficiaries

2025-09-24T11:13:34+00:00

Ensuring safe and convenient access to essential services, including pension retrieval, is crucial in the current digital era. Passwords and PINs are examples of traditional authentication systems that frequently expose people to fraud and identity theft. In order to replace these traditional methods with biometric verification (such as fingerprint and facial recognition), this project suggests a Web Biometric Credentialing System for pension retrieval. The system incorporates Auth0 for secure identity and management of sessions and WebAuthn API for biometric authentication. This method greatly enhances security and user experience by enabling pensioners to verify their identity using biometric information. The technology makes sure that only authorized people can access sensitive financial data and, after successful verification, enables pensioners to safely retrieve their pension amounts. By lowering fraud, eliminating unwanted access, and streamlining the authentication procedure, the suggested solution improves security.