A deep density based and self-determining clustering approach to label unknown traffic

Analyzing non-labeled data is a major concern in the field of intrusion detection as the attack clusters are continuously evolving which are unknown for the system. Many studies have been conducted on different techniques such as clustering to solve this issue. Consequently, in this paper the clustering techniques are applied based on the packets’ similarity to categorize unknown traffic. After clustering is done by the proposed architecture, the security investigator analyzes one packet from each cluster (instead of thousands of packets) and generalize the result of analysis to all packets belonging to the cluster. The proposed architecture, namely Associated Density Based Clustering (ADBC) applies multiple unsupervised algorithms and a co-association matrix to detect attack clusters of any shape as long as they have density-connected elements. Furthermore, the architecture automatically determines the best number of clusters in order to categorize non-labeled data. The performance of proposed architecture is evaluated based on the various metrics, while its generalization capability is tested with three publicly available datasets.