Leveraging GANs for Financial Fraud Detection: A Paradigm Shift

Abstract

This paper addresses the critical issue of financial fraud detection, emphasizing the shift from traditional credit assessments to leveraging big data and machine learning for enhanced risk assessment. It introduces an innovative model that combines autoencoders with adversarial generative learning to tackle the challenge of sample imbalance without relying on actual fraudulent samples. The model, termed AE-GAN, generates fraudulent samples from normal transaction data, facilitating binary classification. The paper highlights two key contributions: the novel approach to sample generation and the construction of a detection model that not only mitigates class imbalance but also improves detection capabilities through adversarial learning. This research underscores the importance of emerging technologies in bolstering financial institutions' competitiveness in the fintech era.