Kernel group method of data handling: Application to regression problems

Abstract

In this paper, a novel Kernel based Soft Computing hybrid viz., Kernel Group Method of Data Handling (KGMDH) is proposed to solve regression problems. In the proposed KGMDH technique, Kernel trick is employed on the input data in order to get Kernel matrix, which in turn becomes input to GMDH. Several experiments are conducted on five benchmark regression datasets to assess the effectiveness of the proposed technique. The results and a statistical t-test conducted thereof indicate that the proposed KGMDH yields more accurate results than the standalone GMDH in most datasets. This is the significant outcome of the study. © 2012 Springer-Verlag.