Robust Statistical Outlier based Feature Selection Technique for Network Intrusion Detection
K. Nageswara, D. Rajyalakshmi, T. Venkateswara Rao · 2012
Abstract- For the last decade, it has become essential to evaluate machine learning techniques for web based intrusion detection on the KDD Cup 99 data set. Most of the computer security breaches cannot be prevented using access and data flow control techniques. This data set has served well to identify attacks using data mining. Furthermore, selecting the relevant set of attributes for data classification is one of the most significant problems in designing a reliable classifier. Existing C4.5 decision tree technology has a problem in their learning phase to detect automatic relevant attribute selection, while some statistical classification algorithms require the feature subset to be selected in a preprocessing phase. Also, C4.5 algorithm needs strong preprocessing algorithm for numerical attributes in order to improve classifier accuracy in terms of Mean root square error. Irrelevant features in the network attack data may degrade the performance of attack detection in the decision tree classifiers. In this paper, we evaluated the influence of attribute pre-selection using Statistical techniques on real-world kddcup99 data set. Experimental result shows that accuracy of the C4.5 classifier could be improved with the robust pre-selection approach when compare to traditional feature selection techniques.