Research On Intrusion Detection Algorithm Based On Dimension Reduction And Parameter Optimization Of Support Vector Machine

With the growing of network technology and attack methods, intrusion detection system gradually aroused the concern of the broad scholars, becoming the important subject in network security research of the current. In numerous intrusion detection method, researchers have found that the support vector machine method, there are many advantages to the field of intrusion detection, and the research on intrusion detection algorithm based on support vector machine has also very important significance.In this article, the performance improvement of intrusion detection algorithm based on support vector machine is deeply studied from three aspects of the feature dimension reduction of the intrusion detection data, parameter optimization of support vector machine and the construction of intrusion detection model.Firstly, kernel principal component analysis algorithm for feature dimension reduction is expounded in this article, and the influence of the penalty parameter and kernel parameter of support vector machine on the classification performance is analyzed, meanwhile, the background and thought of using particle swarm optimization algorithm to optimize parameters of support vector machine are illustrated.Secondly, aiming at the problem of high dimension of intrusion data set, a kernel principle component analysis algorithm based on Relief F algorithm and sample selection is proposed. The algorithm use Relief F algorithm for feature selection, and the kernel principal component analysis algorithm is implemented for each group of samples after feature selection, the samples of each group are screened and filtered by the first two principle components chosen from the implementation results of algorithm, and the kernel principal component analysis algorithm is executed again for the screened samples, and then the final principle components are extracted.Thirdly, aiming at the problem of influence of SVM parameters on the classification performance, a particle swarm optimization algorithm based on synchronization optimization of velocity and displacement is proposed. In order to achieve the purpose of optimizing velocity and displacement of particle, the algorithm is based on the classification of particle population model, by introducing two variables of evolution degree and aggregation degree of particles to achieve the dynamic adjustment of inertia weight, learning factors and time factor, making the algorithm that we proposed can find the optimal support vector machine parameters more quickly.Last, based on the effective of the proposed two kinds of the improved algorithms in the article is verified, an intrusion detection model based on support vector machine is reestablished in this article, and the performance of the model with other intrusion detection model is compared in the MATLAB environment.