Discriminant analysis in pairwise kernel learning for SVM classification

Abstract

Multiple kernel learning arises when different types of kernels are employed simultaneously. In particular, in the situation that the data are from heterogeneous sources. In this study, we developed a new framework for determining the coefficients in learning pairwise kernels for classification in Support Vector Machines (SVM). The effectiveness of the proposed method was then demonstrated through the prediction of self-renewal and pluripotency mESCs stemness membership genes. It was also tested on the power of discrimination in DNA repair gene data. The promising formulation in learning coefficients for pairwise kernel learning was shown via experimental evaluation. This may provide a novel perspective for kernel learning in future applications.