Deep Learning Based Network Intrusion Hybrid Detection Method

• 1、Beijing University of Posts and Telecommunications,School of Cyberspace Security, Beijing 100876

Abstract：In order to address the problems about the huge, complex and high dimentional network data in network intrusion detection, a deep learning based hybrid method for network intrusion detection is proposed in this paper, with the combination of supervised and unsupervised learning methods. First, used sparse stacked auto-encoder to extract features, which had better performance in classification. Second, combined the original features and extracted features to construct a more comprehensive feature set, then used deep neural network to furter select features and classificate the data. At the same time, some elements, which is relative with the discriminating of new attack typewere added to detect the new attack types. Lastly, NSL_KDD as a common public data set was used to evaluate the performace of the propose method from two aspects, including the effect elements of model performance and the performance comparation between different methods. The results show that the proposed method has better performance in discriminating the attackts from normal actions, compared with other traditional methods, especially in detecting new attack types.

Keywords： Information safety Deep learning Network Intrusion Dectection Sparse Stacked Auto-encoder Deep Neural Network K-means clustering