Research On Mixed Kernel Function Based On Support Vector Machine

Support vector machine(SVM)is one of machine learning algorithm which based on statistical learning theory.Because of its rigorous mathematical derivation and simple easy-to-understand algorithm structure,SVM are tends to be with full completeness in linear classification theory,but in terms of the selection of kernel function and its parameters are still no definite conclusion can be used directly.This paper mainly studies the parameter selection of radial basis kernel function(RBF kernel function),Poly kernel function(Poly kernel function)and the improved RBF kernel function--N_RBF kernel function,and deeply studies the mixed form of the above kernel functions.In order to solve the problem of nonlinear classification of support vector machine,this paper uses the powerful ability of kernel function for feature representation and nonlinear classification,introduce the basic properties of RBF kernel function and Poly kernel function in detail and explain why it has a good classification property as a single kernel function.The N_RBF kernels based on scalability are introduced to study and study the properties of N_RBF kernels.The First method of mixed kernel function based on support vector machine is proposed to solve the nonlinear classification problem.The RBF kernel function and the Poly kernel function are mixed by the linear weighted sum method to obtain a new mixed kernel function--RBF_Poly kernel function,which fully utilizes the advantages of the function as a single kernel function.The experimental results have shown that the new proposed RBF_Poly kernel function not only can effectively improve the prediction accuracy,but also has strong generalization performance when faced with diversiform data.The Second method of mixed kernel function based on support vector machine is proposed to solve the nonlinear classification problem.The N_RBF kernel function and the Poly kernel function are mixed by linear weighted sum method to obtain another new mixed kernel function--N_RBF kernel function,which can ensure that the expansion ratio is not less than 1,which overcomes the disadvantage that the RBF kernel function's classification performance decline sharply when the expansion ratio is less than 0.5.Simulation experiments have shown that the N_RBF_Poly kernel function has better classification effect than single kernel function.