Research On The Improvement Of Intrusion Detection Performance Based On SVM

With the development of computer network technology, people's lives and learning methods have undergone tremendous changes .It brings people to the information technology and network-oriented era, at the same time, the problems of network security become increasingly prominent. Intrusion Detection as an important part of information security research has aroused broad attention at home and abroad. Traditional intrusion detection technology has disadvantages like false, omitted and low timeliness, Therefore, this research focuses on how to improve the performance of intrusion detection systems, concrete work as follows:1)We briefly describe the concepts, categories and testing processes of intrusion detection system, points out the problems that the IDS need to solve at present and prospects the future trends; describes the theoretical basis of Support Vector Machine, summarizes many advantages that SVM can be applied to the aspects of intrusion detection; at the same time the paper designs a SVM-based network intrusion detection model, including modular design, parallel detecting, in order to improve overall detection rate of the system.2)We briefly introduces the principle and realization process of the Standard particle swarm optimization algorithm, because of the defect that kernel function parameter values depend on the experience,it proposes an improved particle swarm optimization algorithm, to optimize the kernel parameterσand the penalty factor C .Through the Matlab simulating, it has showed that this revised algorithm improves the SVM-based detection performance of IDS. 3)Applying the improved PSO algorithm to feature selection, searching for the best key features of various types of attacks, eliminating redundant information between the data, it can improve the detection performance of the system. Through the experiments it shows that this optimization scheme is feasible.4)Advances a joint optimization model, optimizes the selection problem of the kernel parameterσ,the penalty factor C and the 41-dimensional feature vector synchronously by the improved PSO algorithm. This optimization program is proved better than separately executing parameter optimization and feature selection through the simulation experiments.