Support Vector Machine for Intrusion Detection Based on LSI Feature Selection
Qing Yang, Fangmin Li · 2006
Data mining as a novel approach has been widely applied to intrusion detection, selecting an appropriate representation to extract the most significant feature is very important, and the algorithm of pattern classification is also crucial. This paper describes a new support vector machine (SVM) for anomaly intrusion detection idea based on latent semantic indexing (LSI). The proposed method can generate features in LSI method by singular value decomposition (SVD). SVD as preprocessing step can reduce the dimensionality and remove the noise in the raw data matrix. The computation complexity is also greatly degraded. SVM has been proved that have a good performance for classification. We performed SVM on the new feature space obtained by SVD, the redial basis function (RBF) is chosen as our kernel function. Our experiments performed on PARPA'98 BSM data set show that our approach can lead to a higher detection rate and a lower false positive rate, it is an effective and efficient methods. In particular, LSI technique can lead to a greatly reduction of the computation complexity and CPU-time