Classification Method Based On Support Vector Machine Network Vulnerabilities

With the development of computer technology, the problem of network security received more and more attention. The existence of network vulnerability is one of the causes which affect the network security. How to classify for vulnerability canonical and reasonable is very important. And that support vector machine is a good tool for classification. Support Vector Machine is a machine learning method based on statistical learning theory. Its most major characteristic is to enhance the learning machine to exude the ability as far as possible according to the Vapnik structure risk smallest principle, namely to obtain the small error by the limited training sample collection to be able to guarantee maintaining the small error to the independent test collection. Moreover, stemming from which the support vector algorithm is a raised optimized question, the partial optimal solution also is the overall situation optimized solution, this is other study algorithms does not compare. Therefore, this paper has proposed an improved classification method of network vulnerability based on multi-class support vector machine.First of all, this paper introduced the conception of network vulnerability and the based theory of support vector machine. Secondly, the multi-class classification methods are summarized including one-against-rest, one-against-one and decision directed acyclic graph support vector machine, and their advantage,disadvantage and capability are compared. The disadvantages of the existing methods are analyzed and compared in this paper. To solve these problems, this paper proposed a new arithmetic of multi-class support vector machine based on binary tree. This arithmetic combine the shortest distance of clustering analysis and the character of network vulnerability to construct the binary tree, shortened the time of vulnerability taxonomy. Follow, for the sake of increase the precision of vulnerability taxonomy, this paper used a method of weighted for those feature that have important function in the course of vulnerability taxonomy. Finally, experiments have been made on vulnerability database, the feature data are preprocessed by means of hash table. Results from experiment indicate the improved methods not only shortened the time, but also improve the precision.