Credit Card Fraud Detection Using Sparse Autoencoder and Generative Adversarial Network

Current credit card detection methods usually utilize the idea of classification, requiring a balanced training dataset which should contain both positive and negative samples. However, we often get highly skewed datasets with very few frauds. In this paper, we want to apply deep learning techniques to help handle this situation. We firstly use sparse autoencoder (SAE) to obtain representations of normal transactions and then train a generative adversarial network (GAN) with these representations. Finally, we combine the SAE and the discriminator of GAN and apply them to detect whether a transaction is genuine or fraud. The experimental results show that our solution outperforms the other state-of-the-art one-class methods.