High Performance WGAN-GP based Multiple-category Network Anomaly Classification System

Due to the increasing of smart devices, the detection of anomalous traffic on Internet is getting more essential. Many previous intrusion detection studies which focused on the classification between normal or anomaly events can be used to enhance the system security by launching alarms as the intrusions being detected. Although many intrusion detection systems which has been developed can achieve high detection rates, they are still difficult to perform well on some attacks that have never been seen before. In this paper, the performance of multiple-category classification on NSL-KDD dataset is evaluated using Wasserstein Generative Adversarial Network - gradient penalty (WGAN-GP) to enhance training effectiveness. The experimental result showed that the proposed method obtained the accuracy rates of 88.2% and 80.8% on binary and five category classification problems respectively.