Ensemble and Genetic Algorithm-based Grading Method for Intrusion Detection Systems

Recently, the Internet has developed a vital part of human life for communication. Although internet serves the civilization in a better way, it also poses some serious threats in the form of cybercrime. Intrusion Detection System is a very important tool for network security. However, the Intrusion Detection System suffers from the problem of handling large volumes of data and produces a high false positive rate. In this paper, a novel Grading method of ensemble has been proposed to overcome the limitation of intrusion detection systems. Partial decision tree (PART), RIpple DOwn Rule (RIDOR) learner and J48 decision tree have been used as base classifiers of the Grading classifier. Optimized Genetic Search algorithms have been used for the selection of features. These three base classifiers have been graded using a Random Forest decision tree as a meta-classifier. Experimental results show that the proposed Grading method of classification offers accuracies of 81.3742%, 99.9159% and 99.8023% on testing, training datasets and cross-validation respectively. It is found that the proposed graded classifier outperforms its base classifiers and existing hybrid intrusion detection system in terms of accuracy, false positive rate and model building time. The genetic algorithm helped to reduce the training time of the proposed Graded classifier by selecting suitable features in the training dataset. In future works, the plan will be to implement a real-time intrusion detection system using real data networks.