Ensemble of Decision Trees for Intrusion Detection System

Abstract

Now-a-days, Internet is playing vital role to change economic, political and social structure positively. In business transaction, the enormous assistance of Internet have stemmed in increased number of users and subsequently intruders. Intrusion Detection System detects intruders in networks. Using traditional approaches of intrusion detection, it is actual difficult to analyze packets in network. Development of Intrusion detection system by using ensemble method is leading to faster and enhance accurate detection rate. In this paper, decision trees based an ensemble classifier has proposed for detection of intrusion in network. The key aim of this research is to develop an ensemble to enhance accuracy of network intrusion detection on testing data set. Three decision trees have used as a base classifier. Because the decision trees are modest in environment and produce simple rules in if-then form. For building and testing the proposed ensemble based classifier, NSL-KDD dataset have used. The novelty of this research work is that ensemble of fast decision trees have combined together which provided very high accuracy. Experimental results shows that the proposed ensemble classifier beats its base classifiers and other existing ensemble classifiers on test dataset. It is also observed that the proposed ensemble classifier offers improved classification accuracy than Random forest and AdaBoost on test data-set. The proposed ensemble classifier also offers better accuracy than existing classifiers on training data-set. The proposed ensemble classifier also provide higher accuracy than classifier proposed in literature on 10-fold cross validation. Overall, the proposed ensemble based classifier beats standard ensemble classifiers and existing classifiers.