Nonlinear regression and multiclass classification via regularized radial basis function networks

We consider the problem of constructing nonlinear regression and multiclass classification models, using radial basis function networks with the help of the technique of regularization. Crucial issues in the model building process are the construction of the basis functions and also the choices of the number of basis functions and a regularization parameter. In order to choose the adjusted parameters, we use model selection and evaluation criteria. We investigate the properties of nonlinear modeling strategies based on radial basis function networks and the performance of model selection criteria from a predictive point of view.