An Ensemble Classifier Approach on Different Feature Selection Methods for Intrusion Detection

Knowing a day's monitoring and analyzing events of network for intrusion detection system is becoming a major task. Intrusion detection system (IDS) is an essential element to detect, identify, and track the attacks. Network attacks are divided into four classes like DoS, Probe, R2L, and U2R. In this paper, ensemble techniques like AdaBoost, Bagging, and Stacking are discussed which helps to build IDS. Ensemble technique is used by combining several machine learning algorithms. Selection of features is one of the important stages in intrusion detection model. Some feature selection methods like Cfs, Chi-square, SU, Gain Ratio, Info Gain, and OneR are used in this paper with suitable search technique to select the relevant features. The selected features are applied on AdaBoost, Bagging, and Stacking with J48 as a base classifier and along with that J48 and PART are used as single classifies. Finally, results are shown that the use of AdaBoost improves the classification accuracy. Experiments and evaluation of the approaches are performed in WEKA data mining tool by using benchmark dataset NSL-KDD '99.