On the Efficiency of Support Vector Classifiers for intrusion detection

We implement multi-class SVMs (by one-versus-rest, one-versus-rest method and a new Decision Tree (DT) SVM) for intrusion detection. None of these methods show advantages over two-class method wherever in detection accuracy or time cost in our experiments. We also apply a support vector (SV) reduction algorithm and find that it decreases the training time dramatically while improves the detection rate.