Credit Card Fraud Detection Using Adversarial Learning
Jitong Geng, Bin Zhang · 2023
In recent years, with the proliferation of the internet and e-commerce, the user base for credit cards has witnessed a continuous surge. However, the presence of credit card fraud has resulted in immeasurable losses for users, merchants, and financial institutions. Contemporary practices in fraud detection primarily rely on classification methods such as CNN, LSTM, and DNN. Nonetheless, these approaches predominantly consider the utilization of original features and exhibit suboptimal performance when confronted with imbalanced datasets. Moreover, they necessitate substantial volumes of annotated data for effective training. This paper introduces an unsupervised anomaly detection network that leverages dual adversarial learning for credit card fraud detection. In contrast to conventional anomaly detection methodologies, our approach emphasizes the simultaneous consideration of both original and intermediate features. Experimental results conducted on the European cardholder dataset demonstrate the superior efficacy of our approach, with an achieved accuracy of 0.9224, F1 score of 0.9208, and MCC of 0.8456, surpassing existing fraud detection techniques.