Support Vector Machine (svm) Parameter Selection Method In Intrusion Detection Research

With the development and popularization of the network, the security of network and information security has been increasingly concerned. Intrusion Detection (ID) is one of the indispensable technologies in network security, and has become a new hotspot for the network security technology in recent years. Support vector machine (SVM) is a machine learning method based on statistical theory which is apt to deal with the problems of small sample, nonlinear and high dimension. The application of SVM to intrusion detection can get a better detection performance.The intrusion detection method based on SVM was studied mainly on the parameter selection of support vector machine in intrusion detection. In order to design an intrusion detection classifier based on SVM with good learning ability and generalization ability, the effects of two parameters of the SVM-the penalty factor and the kernel parameters on the classification performance, as well as the design criteria of intrusion detection classifier based on SVM were discussed, and a recursive way of support vector machine parameters selection was proposed. It simplifies the process of parameter selection through looking for the rules of the variation of the penalty factor and kernel parameter in a specific data space which are used to update them and their different effects on the classification performance. At last, a algorithm of parameters selectiong of SVM classification algorithm based on the Bayesian framework was investigated. Bayesian inference guidelines were applied to accurately explain the training of the support vector machine classifier. The selection of the penalty factor and kenerl parameters of the support vector machine classification algorithm based on the Bayesian framework was untaken, and applied to intrusion detection.The proposed algorithm of parameters selectiong of SVM classification algorithm was used in the intrusion detection data set of KDD Cup99, and was compared with the traditional method based on support vector machines. The results showed that the method got a high detection rate and a reduced time of training model, and that the classification algorism for support vector machines based on the Bayesian framework had a great advantage in the case of small samples.