Algorithm of multi-category SVM incremental learning in application of intrusion detection

This paper proposed a new algorithm of multi-category SVM incremental learning by analyzing the distribution characteristics of the intrusion detection data. Samples used in learning were selected by measuring the distance between samples and their class-centers, and they are just those samples which will most possibly be the SVs in incremental learning. By several binary-class hyper-planes, the zones of the inhomogeneous samples are divided, and the multi-category incremental learning is realized. Using this algorithm, the quantity of training samples is reduced while high detection rate is ensured at the same time. The test of the algorithm is based on the KDDCUP99 dataset. The testing result proved that the time complexity and the space complexity of incremental learning can be both reduced effectively while the accuracy won´t decrease.