CNN-CGAN: A New Approach for Intrusion Detection Based on Generative Adversarial Networks

With malicious traffic occurring all the time, network intrusion detection remains a critical task. However, data imbalance in the network and the increasing number of unknown attack types make the detection difficult. Therefore, in this study, we propose an anomalous traffic detection method named CNN-CGAN, which achieves data equalization by a modified generative adversarial network (GAN) and uses an convolutional neural network (CNN) as the detection model. First, chi-square test is used to extract various types of features from network attack data to accelerate the convergence of the model. Then, we use a improved generative adversarial network to generate data with similar distribution to the small sample data to complete the data equalization. Finally, CNN effectively extracts data features for attack detection and classification. Experiments on the network security dataset NSL-KDD prove that the CNN-GAN model in this paper outperforms the classical detection models in performance indicators such as F1 score, precision and recall. In addition to this, the detection rate of unknown attacks and attack types is also higher with fewer samples.