An information-geometrical method for improving the performance of support vector machine classifiers

The performance of support vector machine (SVM) largely depends on the kernel. There have been no theories concerning how to choose a good kernel in a data-dependent way. As a first step to this important problem, we propose an information-geometrical method of modifying a kernel function to improve the performance of a SVM classifier. The idea is to enlarge the spatial resolution around the separating boundary surface by a conformal mapping. We gave examples of modifying Gaussian radial basis function kernels. Stability of such processes is also known. Simulation results for both artificial and real data turns out to support our idea.