An Intrusion Detection Method based on PCA and ResNet

Abstract

Network Intrusion Detection System(NIDS) is a useful tool to ensure the confidentiality, integrity and availability of Network Information System. Machine learning techniques have been applied to many industrial fields, among which deep learning techniques are the most prominent. To deal with malicious activities in network, an accurate and fast machine learning based intrusion detection method is required. In this paper we propose a fast and accurate intrusion detection method based on PCA and Residual Network(ResNet). PCA is used for dimensionality reduction. Then we design a ResNet and study the impact of number of principal components k and size of time-related parameter time-step t to the classification accuracy of ResNet on UNSW_NBI5 Dataset. To select proper k and t a fast optimization strategy is introduced. Experimental result showed that the accuracy of binary-classification of our method reaches over 98% with a false alarm rate less than 1.8% and less than 400s in model training. For multi-classification, the accuracy reaches over 86%.