Study On Class And Sample Weighted Support Vector Machine And Its Application In Intrusion Detection

Since the creation of the 1999 KDD Cup network intrusion data set, several machine learning approaches to intrusion detection task have been found to be successful. Support Vector Machine (SVM) is developed with the research of Statistical Learning Theory (SLT) and it is a classification and prediction algorithm based on Structural Risk Minimum (SRM) Theory. SVM Theory lies on SLT's VC theory and SRM theory, and compromises limited sample information and the complexity of model so as to get the maximum generality, it is a machine study method with good performance when the sample size is small.Given this feature of SVM, this paper applies SVM algorithm into the network intrusion detection. In the application of network classification using SVM, the use of training sets with uneven class sizes results in classification biases towards the class with the large training size. Shu-Xin Du proposed an improved approach weighted SVM to solve this problem. Weighted support vector machines for classification where penalty of misclassification for each training sample is different, and then the classification accuracy for the class with small training size is improved, and overcomes the drawback which standard support vector machine algorithm can not deal with this sample flexibly, and improves the generalizing ability with given less prior knowledge. However, Weighted SVM considers different penalty parameters about class only and ignores the differences of importance from different samples. In this paper, we introduced class and sample weighted factors respectively and propose a new method, namely, Class and Sample Weighted C-Support Vector Machine (CSWC-SVM) to solve the problem of classification biases towards the class with the large training size. Furthermore we proposed a network intrusion detection model based on algorithm of CSWC-SVM.In this paper, C language to code the data pretreatment is used, and uses MATLAB language to code CSWC-SVM, KDD Cup 1999 Data intrusion detection test data set as the training and testing data. This paper applies CSWC-SVM algorithm into the network intrusion detection.Then the experiments'results express this method has better detection rate, overall accuracy, has robust false positive rate and false negative ratio with less training sample size, and prove it is effective and efficient.