Anomaly Detection by Clustering in the Network
Feng Guo, Yingzhen Yang, Lian Duan · 2009
Intrusions impose serious security threat to network environment, so it is necessary to detect and cope with them. Many intrusion detection methods focus on signature detection, where models are built to recognize known attacks. However, signature detection, limited by its nature, cannot detect novel attacks. New intrusion types, of which detection systems may not even be aware, are difficult to detect. Anomaly detection focuses on modeling the normal behavior and identifying significant deviations, which could be novel attacks. In this paper we present a clustering algorithm to identify outliers. It performs clustering on feature vectors collected from the network and can automatically detect new types of intrusions without need of manual classification of training data. Experimental results show that our system achieves a satisfactory intrusions detection rate while keeping the false positive rate reasonably low.