The determination of kernel function parameter initial value of the KPCA based on adaptive contribution rate

Abstract

In this article, we start from the samples and the kernel principal component contribution rate adaptive point of view. Our aim is to research that how to choose the initial parameters of the mixed kernel function in comprehensive evaluation of KPCA. To some extent, it will overcome the shortcomings that the kernel function parameter is difficult to determine, and it provides an effective method that KPCA applies to determine the kernel function initial parameter of the multi-indexes comprehensive evaluation. We believe that it will be more widely applied to feature extraction with its advantages appear in specific issues, and makes the evaluation method more and more scientific and effective.