Application of Weighted Support Vector Machines to Network Intrusion Detection

摘要: Support Vector Machines(SVMs) have succeeded in many classification fields. Some researchers tried to apply SVMs Intrusion Detection recently and got desirable results. By analyzing C-SVM theoretically experimentally, we found that had some properties which showed was not most suitable for Network Detection. First, has different error rates on classes if the sizes of training are uneven. Second, is over-dependent every sample, even samples duplicated. Third, does make a difference between samples. According these characteristics fact size network normal data always much larger than intrusion importance attack from each other, an extended C-SVM, termed weighted proposed this paper. Weighed introduces two parameters, class weights sample weights. Class used adjust false negative rate positive class. And emphasize Experiments Weighted more effective detection systems.