Research on Parameter Optimization of the Boolean Kernel Function

Abstract

It is significant to choose super parameter of the kernel function in non-liner SVM for the constructed classifier. As to the boolean kernel function, the research is comparatively less than other kernel function in the internal and international country, and the selection of its parameter is mainly through handwork. This paper researched and analyzed some main measure to choosing parameter of kernel function, discussed the optimizing principle of parameter, which to minimize the RM bound. In the paper, it proposed an abbreviated algorithm by adopting constant iterative step length to optimize the parameter, aiming at super parameter of the boolean kernel function KMDNF, which implemented automatic selection.