A comparison of sparse kernel principal component analysis methods

Abstract

Presents a comparative study of a group of methods based on kernels which attempt to identify only the most significant cases with which to create the nonlinear feature space. Kernels were originally derived in the context of support vector machines, which identify the smallest number of data points necessary to solve a particular problem (e.g. regression or classification). We use extensions of kernel principal component analysis to identify the optimal cases to create a sparse representation in feature space. The efficiency of the kernel models is compared on an oceanographic problem.