A novel three-phase algorithm for RBF neural network center selection

Abstract

In this paper, we propose a new method for selecting RBF centers. The strength of our method is to determine the number and the locations of RBF centers automatically without any priori assumption about the number of centers. The proposed method consists of three phases. The first phase is to partition the input patterns into the several subsets according to; their output labels. In the second and third phase, the number and the locations of RBF centers are determined using bisection algorithm and weighted mean centering. These second and third phase are iteratively, repeated until to reach the goal error. The proposed method is applied to several benchmark data sets. The numerical results show that our method is robust and efficient for determining the number and the locations of centers.