Incremental SVM algorithm to intrusion detection base on boundary areas

The B-ISVM method based on a fast incremental SVM is proposed in this paper for the low rate of intrusion detection and the slow detection speed of standard SVM method. The first step is to identify boundary areas, train screened boundary areas samples in order to construct the initial classification hyperplane. Then, the support vector is extracted effectively according to filtering factor. Finally, the construction of the incremental SVM classifier is completed through incremental learning based on KKT conditions. The experiment results show that the method could achieve the higher rate of intrusion detection and faster detection speed. Thus the proposed method is overall superior to the standard SVM and ISVM method in terms of classification performance.