The Study On Selecting Parameters Of SVM Using Genetic Algorithm

Now Support Vector Machine (SVM) has been widely used in machine study area, whose parameter selection is still a critical issue. Genetic Algorithm(GA) is a kind of self-adapt global optimization probability searching algorithm by simulating creatures genetics and evaluation in natural environment. Using genetic algorithms, this article focuses on model selection of SV Classification and Clustering parameters. This paper also focuses on solving the two problems, one of which is GA selecting the SV Classi-fication parameter based on the inter-cluster distance; the other is selecting SV Cluster-ing parameter using GA. The aim of this paper is to avoid unsatisfactory results or fail-ure of the experiments due to pre-setting the parameters of classification and clustering. Main results of this study are:Firstly, using GA, the algorithm of selecting the Gauss kernel function parameters and the penalty factor C of SV Classification, based on the inter-cluster distance is respectively proposed. Next, the algorithm is compared with SV Classification algo-rithm whose parameter is selected artificially and SV Classification algorithm whose parameter is selected by GA. Some simulation experiments are made.Secondly, using GA, the algorithm of selecting the Gauss kernel function parame-ters and the penalty factor C of SV Classification, based on the inter-cluster distance is proposed. Next, this algorithm is compared with the algorithm which was proposed in the first model in the simulation experiments.Thirdly, using hybrid genetic algorithm(HGA), the algorithm of selecting the ker-nel function parameters and the penalty factor C of SV Classification, based on the inter-cluster distance is proposed at the same time. Next, this algorithm is compared with the other algorithm in simulation experiments.Finally, The application of GA is studied in SV Clustering. Using GA, the algo-rithm of selecting SV Clustering parameters is proposed and using HGA, the algorithm of selecting the kernel function, the kernel parameters and the penalty factor C of SV Clustering is proposed at the same time. Next, above algorithms simulated experiments on different datas.