DAREnsemble: Decision tree and rule learner based ensemble for network intrusion detection system

Abstract

The Intrusion detection system is a network security application which detects anomalies and attackers. Therefore, there is a need of devising and developing a robust and reliable intrusion detection system. Different techniques of machine learning have been used to implement intrusion detection systems. Recently, ensemble of different classifiers is widely used to implement it. In ensemble method, the appropriate selection of base classifiers is a very important process. In this paper, the issues of base classifiers selection are discussed. The main goal of this experimental work is to find out the appropriate base classifiers for ensemble classifier. The best set of base classifier and the best combination rules are identified to build ensemble classifier. A new architecture, DAREnsemble, have proposed for intrusion detection system that consists of unstable base classifiers. DAREnsemble is formulated by combining the advantages of rule learners and decision trees. The performance of the proposed ensemble based classifier for intrusion detection system has evaluated in terms of false positives, root mean squared error and classification accuracy. The experimental results show that the proposed ensemble classifier for intrusion detection system exhibits lowest false positive rate with higher classification accuracy at the expense of model building time and increased complexity.