SVM ensemble intrusion detection model based on Rough Set feature reduct

To address the problem of low accuracy and high false alarm rate in network intrusion detection system, an intrusion detection model of SVM ensemble using rough set feature reduct is presented. Utilizing the character that rough set algorithm can keep the discernibility of original dataset after reduction, the reducts of the original dataset are calculated and used to train individual SVM classifier for ensemble, which increase the diversity between individual classifiers, and consequently, increase the probability of detection accuracy improving. To validate the effectiveness of the proposed method, simulation experiments are performed based on the KDD 99 dataset. During the process of the experiments, two arguments, the sample number and the base classification number, are discussed to test their effect on the final result. And then detection performance comparison among the SVM using all samples, SVM-bagging ensemble and rough set based SVM-bagging are performed. The results show that the Rough Set based SVM-bagging is a promised ensemble method owning to its high diversity, high detection accuracy and faster speed in intrusion detection.