An empirical comparison of training algorithms for radial basis functions

Abstract

In this paper we present a review and comparison of five different algorithms for training a RBF network. The algorithms are compared using nine databases. Our results show that the simplest algorithm, k-means clustering, may be the best alternative. The results of RBF are also compared with the results of Multilayer Feedforward with Backpropagation, the performance of a RBF network trained with k-means clustering is slightly better and the computational cost considerably lower. So we think that RBF may be a better alternative. © Springer-Verlag Berlin Heidelberg 2003.