DACNN: Deep Autoencoding Convolutional Neural Network in Network Intrusion Detection

Abstract

In recent years, the Internet has become the main driving force of global economic growth. However, the everchanging network security situation is not optimistic. With the development of network technology, the threat of network attacks to key network nodes has seriously affected the security of the entire network situation. In this article, We propose an effective intrusion detection method based on deep autoencoding convolutional neural network: DACNN. First, we trained multiple autoencoders with good performance. Then combine the dimensionality reduction representation of flow information obtained by the autoencoders with the original information, so that convolutional neural networks can use this kind of mixed information that contains the original data and the dimensionality reduction information processed by the autoencoder to analyze the network traffic information. The DACNN has better recognition performance in the field of intrusion detection than ordinary convolutional networks. Tests on multiple datasets show excellent intrusion detection performance:ACC 0.935 in KDDCUP and F1 0.98 in IDS-2017.