A Deep Learning and Resampling Approach to Credit Card Fraud Detection

Abstract

Credit card fraud presents a major challenge for both financial institutions and individuals, with fraudsters employing increasingly complex methods. Traditional machine learning methods often fail to accurately identify fraudulent transactions due to the issue of imbalanced datasets in credit card transactions. In response, we introduce an approach that combines deep learning with resampling techniques to improve the detection of credit card fraud. This study evaluates various machine learning models and tests different oversampling methods. Our results show that combining Generative Adversarial Networks (GANs) with the random forest model significantly outperforms other methods in detecting credit card fraud. This method effectively mitigates the problem of class imbalance, thereby increasing both the accuracy and the efficiency of fraud detection systems. This suggests promising enhancements for the application of fraud detection in real-world scenarios.