Comparison Research of Single Kernel and Multi-kernel Relevance Vector Machine

This paper researches the selection problem of kernel function for Relevance Vector Machine(RVM).Improved Gauss kernel function is proposed.The characteristic of improved Gauss kernel function and normal Gauss kernel function are compared.The improving performance of proposed kernel function is validated.Besides the improving of single kernel function,multi-kernel RVM is researched,by combining local Gaussian kernel and global polynomial kernel,form multi-kernel function,and use it in RVM.Comparison experiments of kinds of kernel functions run on different datasets,and the performance of improved Gauss kernel function and mixture kernel function are validated.