A novel method for improving the classification capability of radial basis probabilistic neural network classifiers

This paper proposes a novel method for improving the classification capability of radial basis probabilistic neural network classifiers. That is, for each pattern class, over one output node, also called class node, are employed to express corresponding input pattern features compared with previous one output node for one pattern class, which will cause the classification reliability and generalization capability to be improved. The experimental results about classifying the parity 3 problem show that such an enhanced classifier network is indeed capable of improving the generalization capability