Title: Kernel methods for regression model based on variable selection

Authors: Sei-ichi Ikeda, Yoshiharu Sato

Addresses: Division of Computer Science, Graduate School of Information Science and Technology, Hokkaido University, 060-0814, Sapporo, Japan. ' Division of Computer Science, Graduate School of Information Science and Technology, Hokkaido University, 060-0814, Sapporo, Japan

Abstract: The aim of this paper is to get a sufficiently smooth regression function using kernel method based on the concept of variable selection for regression model. It is essential point that the procedure does not contain the concept of the regularisation. The criterion of the variable selection is a simple AIC. On the other hand, Fisher|s linear discriminant function (LDF) for two groups is known as the linear regression function which has dichotomous independent variable. Then kernel Fisher|s LDF can be also discussed by the use of variable selection. In this paper, we show that the variable selection method for kernel linear models is also useful and simpler method than several regularisation methods.

Keywords: kernel method; variable selection; regression modelling; discriminant function; smooth regression function; kernel linear models.

DOI: 10.1504/IJKESDP.2009.021984

International Journal of Knowledge Engineering and Soft Data Paradigms, 2009 Vol.1 No.1, pp.49 - 62

Published online: 15 Dec 2008 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article