CREDIT CARD FRAUD DETECTION USING MACHINE LEARNING TECHNIQUES

Abstract

The banks and financial industries are seriously threatened by credit card theft. This research investigates how well different Deep Learning (DL) and Machine Learning (ML) models identify fraudulent transactions. The primary objective is to analyze and compare different strategies for fraud detection to develop a more reliable and accurate decision-making system. The research reviews the challenges in fraud detection and presents solutions by highlighting both established and emerging fraud patterns. In recent years, the rapid increase in online payments through credit cards and UPI has been accompanied by a corresponding rise in fraudulent activities. Fraudsters employ a wide range of techniques such as card theft, swapping, phishing, and large-scale data breaches to obtain sensitive card information. Due to the high volume of genuine transactions and the limited number of fraud cases, the transaction datasets are often extremely imbalanced, leading to challenges such as model bias, poor generalization, and misleading performance metrics. In order to balance the dataset, the research uses Generative Adversarial Networks (GANs) to create artificial samples of the minority (fraudulent) class. The Synthetic Minority Oversampling Technique (SMOTE) is used to compare the performance of GANs. Experimental results show that GAN-based resampling yields better classification performance, particularly in terms of F1- score. Given its ability to manage high-dimensional and imbalanced data, GAN proves to be a powerful tool for financial fraud detection. The European Cardholders 2013 dataset is used to evaluate the models. This study demonstrates that incorporating GANs can significantly enhance fraud detection systems and emphasizes the need for continuous innovation to address evolving fraud patterns and improve financial security.