Research About Network Intrusion Detection Algorithm Based On Confidence Machine

Along with the wide application of computer and Internet, the security problem about systems or network becomes more and more important. Intrusion Detection Systems are complementarities to security mechanism of network and computer, and enhance the protection depth of security.Support Vector Machine (SVM) is an algorithm of machine learning, which is based on the Statistic Learning Theory (SLT). Compared with the statistics which researches the rule got by samples or requests the number of samples go infinite, SVM puts more attention to the information provided by the samples, and fits to solving the small sample problem. The Nearest Neighbour method is one of the most important methods on no-parameter pattern recognition. With the feature of network data, based on the SVM and NN method, the paper introduces the confidence and the credibility measures, and puts forward a new classification algorithm Nearest Neighbour Confidence Support Vector Machine(NN-CSVM), applying the algorithm in the stage of network intrusion detection. The algorithm improves the classification veracity by improving accuracy of classifying samples about optimal hyperplane. Further more, the algorithm decreases the number of representation dots in order to improve classification velocity. The confidence and credibility measures are import to response component of network intrusion detection system.Java and Weka3.5 platform are used to code the algorithm program and DARPA intrusion detection test data set is used as the training and testing data. Experiment results show that the NN-CSVM algorithm is more effective and more accuracy than the pure statistical method.