A Vulnerability Classifier Based On SVM Optimized By GA-PSO

Information network and computer has become an indispensable partof people’s life, study and work, bringing a lot of conveniences. However,because of exploiting vulnerabilities by illegal attacks of hackers, they ledto the emergence of many major network security events. Vulnerabilities arethe chief culprit of network security incidents. In recent years, the researchon vulnerability has been a hot topic in computer and network security.Many studies have shown that most of vulnerabilities are similar in someways. Therefore, how to reasonably and effectively categorize the rapidlyincreased number of vulnerabilities is very important.Support Vector Machine (SVM) is a supervised learning method basedon statistical learning theory. SVM can solve the problem of classification.It applies to small sample, nonlinear and high dimensional patternrecognition etc. However, while solving large sample of data sets, SVM hassome disadvantages, such as be time-consuming and be easy to fall intolocal minimum. In order to make up for the deficiency of SVM, then thispaper introduces genetic algorithm (GA) and particle swarm optimization(PSO) algorithm. Therefore, by doing research on vulnerability, SVM, GAand PSO, this paper has proposed three new improved classificationsmethods of vulnerability based on optimized support vector machine.This paper first introduces some theoretical concepts aboutvulnerabilities and the basic theory of SVM, GA and PSO. Secondly, summarize the merits and demerits of the current commonly usedclassification methods based on SVM. Sum up three points of GA and PSO:combination ways, advantages and disadvantages. Thirdly, In order toimprove the accuracy of vulnerability classification, it is necessary toimprove traditional SVM. There are three ways of optimization: GA, PSOand PSO based on GA. They are proposed that a vulnerability classifierbased on SVM optimized with GA, a vulnerability classifier based on SVMoptimized with PSO and a vulnerability classifier based on SVM optimizedwith GA-PSO. Fourth, build a small database of vulnerability and doclassification for the vulnerability. Finally, the experiment showed that threekinds of proposed classifiers not only shortens the time of classification andalso improve the precision of the vulnerability classification. Theperformance of a vulnerability classifier based on SVM optimized withGA-PSO is better than that of the other two classifiers.