Study on an Improved SVM Model for Intrusion Detection

Abstract

The intrusion detection rate is often affected by the structure parameters of the support vector machine (SVM) model. Improper SVM model design may lead to a low detection precision. To overcome these problems, a new intrusion detection method, based on kernel principal component analysis (KPCA), Particle swarm optimization (PSO), and SVM, is proposed in this chapter. The KPCA was firstly used to obtain the most distinct features of the input data to eliminate the redundant features. Then the PSO was employed to optimize the training procedure of the SVM. Thus, a satisfactory SVM model with good extendable ability was attained. The efficiency of the proposed method was evaluated with the KDD dataset. The results of the experiment demonstrate that the proposed approach outperforms the existing methods, such as PCA-SVM, KPCA-SVM, and SVM with respect to the intrusion detection rate.