Research On Parameter Selection Method For Support Vector Machines

Support Vector Machine(SVM),as a new artificial intelligence technology,has the advantage that it can solve the learning problem when the sample is not enough,and get better results.At present,SVM is a hotspot in artificial intelligence research.Many studies have pointed out that the effect of SVM classification is greatly affected by SVM parameter selection.For the research of SVM classifier,it is usually necessary to consider the optimal selection of kernel parameters and error penalty parameters.At present,the most commonly used parameter selection methods are: grid search,as well as based on genetic algorithms or particle swarm optimization algorithm such as parameter selection.Because there is no unified theoretical guidance,these methods cannot be widely used.In this paper,the optimization effects of the kernel parameters and the error penalty parameters are verified respectively,and then they are used as a parameter to verify the optimization effect.The main work of this paper is as follows.(1)Improved kernel parameter selection methodFor the parameter selection method of SVM based on separation interval,some sample data are neglected in the optimization process,which leads to the unreliability of the training results.A new method for optimizing the kernel parameter selection of SVM is proposed.The new method obtains the maximum value of the distance between each sample data and the center location distance of each class sample data set.At the same time,the independent variable corresponding to the maximum value is selected as the optimum kernel parameter value.(2)Improved error penalty parameter selection methodIn view of the lack of support vector count method to find the optimal error penalty parameter,and according to the non boundary support vector number,characteristics of support vector machine is more stable,the boundary support vectors and non boundary support vector is introduced into the process of optimization error penalty parameter,so that the support vector machine not only has good generalization ability.And the trained model can guarantee its stability.(3)Experimental verificationFirst,through the validation of several different types of data sets,the improved kernel parameter method and the improved penalty parameter method have higher test accuracy than before.At the same time,using both as parameters,the training time is shortened and the testing accuracy is improved.