Radial Basis Function Neural Networks and Principal Component Analysis for Pattern Classification

Radial basis function (RBF) neural networks provide great possibilities for solving signal processing and pattern classification problems. Several algorithms have been proposed for choosing the RBF prototypes and training the network. A supervised learning algorithm based on gradient descent for training RBF neural networks is presented in this paper. This paper also proposes a principal component analysis (PCA) for finding out the number of classes in a pattern classification problem. Simulation results are presented as applied to the Iris classification problem.