A Classified Method Based on Support Vector Machine for Grid Computing Intrusion Detection

Abstract

A novel ID method based on Support Vector Machine (SVM) is proposed to solve the classification problem for the large amount of raw intrusion event dataset of the grid computing environment. A new radial basic function (RBF), based on heterogeneous value difference metric (HVDM) of heterogeneous datasets, is developed. Two different types of SVM, Supervised C_SVM and unsupervised One_Class SVM algorithms with kernel function, are applied to detect the anomaly network connection records. The experimental results of our method on the corpus of data collected by Lincoln Labs at MIT for an intrusion detection system evaluation sponsored by the U.S. Defense Advanced Research Projects Agency (DARPA) shows that the proposed method is feasible and effective. © Springer-Verlag 2004.