Evolutionary Radial Basis Function Network for Classificatory Problems

Classification has been a major problem of study whose application includes speaker recognition, character recognition, etc. In this paper we first adapt the Radial Basis Function Network (RBFN) for classification problems and then use customized Evolutionary Algorithms to evolve the RBFN. The neurons of the RBFN correspond to some class out of the available output classes. Linear addition of only the same class neurons is taken and an additional layer is added that decides the final output on the basis of maximum activation of each class. Evolutionary algorithm has operators jump and add neuron that aid in optimization. Penalty has been used to restrict overgrowth of network. The algorithm was used to solve the problem of detection of PIMA Indian diabetes and gave a recognition rate of 82.37%, which was better than most of the commonly known algorithms in literature.