Based On Support Vector Machine Intrusion Detection Research

With the development of networks, network security becomes more and more prominent, in particular, attacks on computer systems becomes complex and diverse. Network attack detection system, which increases the system and network security's scope of protection, is a kind of supplement of traditional computer security method.SVM is a machine learning algorithm based on the statistical learning theory in the beginning of the 1990s. It was paid more attention to and get success in application of pattern recognition and regression estimation.It combines the Maximal Margin Principle and the Kernel Function Theory. It has a solid theoretical foundation, the ability to promote good and strong nonlinear processing capacity, and can overcome the "Course Dimensionality".This paper has made some attempt on kernel function and kernel parameter selection, includes:This paper proposes an intrusion detection model based on SVM. Since SVM which applies to intrusion detection needs to high training speed, this paper carries out sampling training data to cut down the data scale and raise the training speed.This paper makes use of upper distance of isomerism data collection function HVDM to normalize characteristic data, and proposed the concrete characteristic information normalizing method.Different network characteristics have different impact to classification result. This paper strengthens some characteristics with weighting to find an optimum classified boundary. At the same time we haven't abandoned those affect-weaker characteristics.This paper has selected the evaluating data collection of Defense Advanced Research Projects Agency of USA (DARPA) to test the validity and performance of the congeneric detecting algorithms. And the result indicates that SVM applied to network intrusion detection is effectual.