EBJRV: An Ensemble of Bagging, J48 and Random Committee by Voting for Efficient Classification of Intrusions

Abstract

An effective Intrusion Detection System must be able to classify any ongoing intrusion activity as, ‘abnormal’ with utmost accuracy. The key factor that mainly affects the accuracy of an Intrusion Detection System is the selection of a classification algorithm whose True Positive Rate is the maximum and False Positive Rate is the minimum. An efficient classification algorithm can thus greatly improve the accuracy of the Intrusion Detection System. To ensure the time taken to build the model is the least, Information Gain Feature Selection algorithm is used for ranking the Features. A standard deviation of all the ranks is computed and all the features that are less than the standard deviation value are discarded. This results in the selection of a sub feature set of only 16 out of 41 features available in the data set. When voting of Bagging, J48 and Random Committee techniques for classification is carried out on this reduced feature set, most encouraging accuracy values can be achieved.