A Network Intrusion Detection Approach Based on Asymmetric Convolutional Autoencoder

Network intrusion detection is an important way to protect cyberspace security. However, it still faces many challenges. The network traffic and intrusion behaviors are always very complex and changeable. Deep learning is a potential method for network intrusion detection. In this paper, we first propose an asymmetric convolutional autoencoder (ACAE) for feature learning. Then, we propose a network intrusion detection model by combining asymmetric convolutional autoencoder and random forest. This approach can well combine the advantages of deep learning and shallow learning. Our proposed approach is evaluated on KDD99 and NSL-KDD dataset, and is also compared with other intrusion detection approaches. Our model can effectively improve the classification accuracy of network abnormal traffic. Furthermore, it has strong robustness and scalability.