An Intrusion Detection Approach Based On Analysis Of Cluster Heterogeneity

Abstract

The security of prodigious proportion of variegated information available in computers, connected by a network, can be procured by the employment of suitable Intrusion Detection Systems (IDS). In this paper, a novel intrusion detection approach has been presented that relies on analysing the heterogeneity of clusters formed by reducing the dimensions of data points. The class dominance in each cluster devises the premise of initial classification, which is followed by the appraisal of cluster confidence and cluster spread. The final refined detection is ensured based on the criteria, formulated by these parameters. Experiment conducted on the widely documented KDD Cup '99 dataset infers the efficacy of the approach in terms of high accuracy and low false alarm rates. The robustness and effectiveness is also parametrized by the run time analysis of the system.