Anomaly intrusion detection based on modified SVM

Abstract

A modified SVM multi-classification algorithm integrated with discriminant analysis (D-SVM) was pro-posed,which could solve the problem of low detection accuracy and high false alarm rate caused by unbalanced datasets.For a multi-classification problem could be divided into several binary classification problems,D-SVM could not only have the virtue of high detection accuracy,but also have a low false alarm rate even confronted with unbalanced datasets.Experiments based on KDD99 dataset verify the feasibility and validity of the integrated ap-proach.Results show that when confronted with multi-classification problems,D-SVM could achieve a high detec-tion accuracy and low false alarm rate even when SVM alone fails because of the unbalanced datasets.