Machine Learning Techniques in Credit Card Fraud Detection: A Hybrid Supervised and Unsupervised Approach

Abstract

In the dynamic landscape of financial transactions, the escalating threat of fraudulent activities necessitates cutting-edge solutions for real-time detection. This research introduces an innovative approach utilizing the Kaggle credit card dataset, focusing on comparing the effectiveness of hybrid models versus purely supervised learning models. While traditional models rely solely on supervised learning, this study explores the potential performance gains of integrating unsupervised learning (USL) into supervised learning (SL) frameworks. The core investigation centers on whether unsupervised clustering can enhance pattern recognition in unlabeled data and subsequently improve the performance of supervised models. This research not only evaluates the practical benefits of hybrid methodologies in fraud detection, but also advances real-time analytics through Power BI, aiming to provide a more comprehensive and adaptive solution to emerging financial threats. The algorithms yield an accuracy of 99.75% and a remarkably low underkill rate of 0.20%, demonstrating the effectiveness of integrating human oversight with advanced machine learning techniques.