Rule extraction from differential evolution trained radial basis function network using genetic algorithms

In this paper, we present a GA based methodology for extracting rules from radial basis function neural network trained by differential evolution. Rules are extracted using GATree. Here outputs predicted by the differential evolution trained radial basis function network along with the input variables are fed to the GATree for rule extraction purpose. The performance of the hybrid method was tested on three benchmark datasets namely Iris, Wine and Wisconsin Breast Cancer, using 10-fold cross validation. The rules extracted by the hybrid yielded high accuracies on all datasets.

I. Introduction

Artificial Neural Network (ANN) has been used in several real-time problems. Multilayer Perceptron (MLP) and Radial Basis Function Network (RBF) are the two basic networks which generalize well on classification and regression problems. Training of RBF consists of both unsupervised and supervised schemes in tandem. Unsupervised phase involves K-means clustering and supervised phase consists of linear least squares technique. The parameters of RBF are centers and widths of the clusters in unsupervised phase and weights between the kernel layer and the output layer in the supervised phase [2].