Improvement Research On Kernel Function Of Support Vector Machine For Large-scale Data

Kernel function is the key influence factor on the performance of support vector machine(SVM),but till now,there is still not a systematic theory to direct how to select the kernel function.Large-scale data is widespread in practical problems,but the SVM based on traditional kernel can't satisfy the requirements of this kind of data.To improve the effectiveness of SVM on large-scale data,a new kind of kernel function is constructed.And the most suitable parameter selection algorithm for the new kernel is selected.Based on the constructal theory of kernel,AW kernel function is constructed by combining Askey-Wilson orthogonal polynomial with the simple form of RBF.Aiming at the problems that genetic algorithm(GA)can't make good use of output information and artificial bee colony algorithm(ABCA)is slow at the beginning of search,the fusion algorithm called G-ABCA is gotten by using GA to get nectar source and using ABCA to search the optimization result.The experiments indicate that AW kernel has better effectivity than the common single kernel functions and orthogonal polynomial kernel functions,G-ABCA and ABCA both have better optimization ability in AW kernel than GA,G-ABCA can save more time under a certain condition of accuracy,ABCA is more beneficial to get high accuracy.Using AW kernel function with G-ABCA(or ABCA)can construct SVM model for large-scale data,so as to provide a more effective way for the analysis of large-scale data.