A Rough Set and SVM Based Intrusion Detection Classifier

Support vector machine-based intrusion detection methods are increasingly being researched because it can detect novel attacks. But solving a support vector machine problem is a typical quadratic optimization problem, which is influenced by the feature dimensions and number of training samples. Feature selection or attribution reduction can help reduce the SVM classification time and saving memory space effectively. This paper concerns using rough set for attribution ranking and reducing and using support vector machine for intrusion detection classification. An elicitation attribution reduction algorithm (EARA) based on attribution significance and discernibility matrix is presented and three data discretization algorithms were applied to identify the important attributions. The classification performance of the presented algorithm and classical SVM were compared in accuracy, time, false positive rate, and detection rate. The experiment results show the presented algorithm has ability to reduce the complexity of the structure of the support vector machine, simplify training sets and decrease training time and data storage without obviously sacrificing the detection correctness.