Initializing of an RBF network by a genetic algorithm

In this paper we use a genetic algorithm (GA) for selecting the initial seed points (prototypes, kernels) for a Radial Basis Function (RBF) classifier. The chromosome is directly mapped onto the training set and represents a subset: it contains 1 at the ith position if the ith element of the set is included, and 0, otherwise. Thus the GA serves a condensing technique that can hopefully lead to a small subset which still retains relevant classification information. We propose to use the set corresponding to the best chromosome from the final population as the seed points of the RBF network. Simulated annealing is used to tune the parameters of the radial function without changing kernels location. Experimental results with IRIS and two-spirals data sets are presented.