Multi-Critics Generative Adversarial Networks for Imbalanced Data in Intrusion Detection System
Haofan Wang, Farah I. Kandah · 2024
Network security has always been a research area of significant concern worldwide. The Intrusion Detection System (IDS), as a crucial defensive measure against network attacks, has undergone multiple iterations and evolutions since its inception to adapt to the ever-changing network environment. In this work we propose an imbalance solution to address the increasingly severe challenges in cybersecurity, particularly focusing on the class imbalance problem (long tail distribution) prevalent in network intrusion detection. Utilizing generative adversarial networks (GANs), the model addresses this imbalance by enhancing data quality, which is analyzed using correlation heatmaps and PCA plots. After improving the dataset with GANs, the updated NSL-KDD database is employed for feature extraction using autoencoders, thereby effectively tackling the class imbalance issue to improve detection capabilities. This model is then compared with several traditional GAN models. The results indicate that while the GAN-generated data retains the original database's statistical characteristics, it also addresses the issue of imbalance. Moreover, the subsequent multi-layered processing enables the overall model to more effectively handle various types of attacks.