Ensemble of clustering algorithms for anomaly intrusion detection system

Abstract

Maximizing detection accuracy and minimizing the false alarm rate are two major challenges in the design of an anomaly Intrusion Detection System (IDS). These challenges can be handled by designing an ensemble classifier for detecting all classes of attacks. This is because, single classifier technique fails to achieve acceptable false alarm rate and detection accuracy for all classes of attacks. In ensemble classifier, the output of several algorithms used as predictors for a particular problem are combined to improve the detection accuracy and minimize false alarm rate of the overall system. Therefore, this paper has proposed a new ensemble classifier based on clustering method to address the intrusion detection problem in the network. The clustering techniques combined in the proposed ensemble classifier are KM-GSA, KM-PSO and Fuzzy C-Means (FCM). Experimental results showed an improvement in the detection accuracy for all classes of network traffic i.e., Normal, Probe, DoS, U2R and R2L. Hence, this validates the proposed ensemble classifier.