Optimization of radial basis function networks by genetic algorithms

The effects of a neural network's topology on its performance are well known, yet thequestion of finding optimal configuration remains largely open. In Radial Basis Function(RBF) networks, placement of centers is said to have significant effect on the performance ofnetwork. In this paper, centers and widths of hidden layer are coded in a chromosome andthese two critical parameters are determined with the optimization by Genetic Algorithms(GA). The maximum and minumun limit values of these parameters are defined in thealgorithm whilst centers should be perceived as the number of lines in dataset.This treatise aims to classify the test set with high accuracy while minumum number ofinstance is chosen from the train set. Iris plant, thyroid disease, escherichia coli bacteria, fetus,glass and lens datasets are used to compare the success between GA-RBF and RBF method.The final conclusion is that GA-RBF approach is more effective than RBF and somenumerical results indicate the applicability of the proposed approach, which all simulationsrealized by Matlab 7. Nevertheless, GA took a long training time to achieve these results. Butfor a large number of applications and rough datasets, genetic approach is an attractivesolution for the design of efficient artificial neural networks.Keywords: Genetic algorithms, radial basis functions, selection of centers, Gauss functionspread value