Divergence-driven generative adversarial networks for semi-supervised anomaly detection

Abstract

Anomaly detection is a critical challenge due to the extreme class imbalance between anomalous and normal data instances. This study presents a semi-supervised anomaly detection framework based on GANs, in which the discriminator is repurposed as a direct classifier for anomaly detection. This research advances the application of GANs in anomaly detection by introducing a divergence-driven training paradigm and validating its effectiveness through extensive experimentation with credit card fraud data and Twitter bot accounts.