A bagging approach to network intrusion detection

Abstract

Accompanying the benefits of Internet are various techniques of compromising the integrity and availability of the system connected to it due to flaws in its protocols and software widely entrenched. The presences of these flaws make a secured system a mirage for now, hence the need for intrusion detection system. In this paper, an ensemble approach – Bagging was used on five different machine learning techniques to improve accuracy of classifiers. Machine learning seeks for methods of extracting hidden pattern from data and come up with its own rules based on given data set. The five techniques were made up of two unsupervised (clustering) techniques – Kmeans and Fuzzy Rough C-means, and three supervised (classification) techniques – TreeReduct, LEM2 and Bayesian. Experimental study was carried out on the International Knowledge Discovery and Data Mining Tools Competition (KDD) dataset for benchmarking intrusion detection systems. The results generated from the experiment revealed that ensemble approach performance on the attack types and normal is slightly better or equal to the best performed algorithm on that particular class.

Journal of the Nigerian Association of Mathematical Physics, Volume 15 (November, 2009), pp 379 - 390