Efficient Training of Rbf Networks via the Kurtosis and Skewness Minimization Learning Algorithm

Radical Basis Function (RBF) networks have been widely used in time series prediction because of their simplicity, robustness, good approximation and generalization ability. However, it is still rather difficult to select the number and locations of the hidden units of the RBF network appropriately for a specific time series prediction problem. In this paper, the Generalized RBF networks have been established with Gaussian density functions instead of Gaussian functions for the hidden units. Then, we utilize the kurtosis and skewness minimization criterion as well as the corresponding learning algorithm to select the appropriate number and initial parameters of the hidden units from an input data set automatically for the generalized RBF network on a time series prediction task. It has been demonstrated by the experiments that the generalized RBF network trained with the kurtosis and skewness minimization learning algorithm is feasible and efficient on typical chaos time series.