A Novel Method of Outliers within Data Streams Based on Clustering Evolving Model for Detecting Intrusion Attacks of Unknown Type

It is an important issue to detect the intrusion attacks for the security of network communication. The clustering-based methods usually are proposed to cope with the problem of intrusion detections. However, how to detect the unknown intrusion attacks within stream data has come to be a challenge. In this paper, we consider the intrusion attacks as outliers and propose a novel approach (called DOExMiCluster) based on clustering data stream to detect the outliers of unknown type. The new micro-cluster concept, normalization data technology and k-mean measure are only used to learn the normal sub micro-clusters online till the event that two special micro-clusters are merged and a new micro-cluster is created doesn't appear, and then system recognizes the instances which cannot fall into any micro-clusters as outliers.