Performance evaluation of radial basis function neural networks

Abstract

The work was done to compare the performance of the radial basis function neural networks with that of back propagation neural networks. The comparison was made both in the field of function approximation and pattern recognition. Cosine function and the hermite's polynomial were used for function approximation comparison. For pattern recognition problem, a set of twenty-six English alphabets and another set of ten numeric digits were used. The various comparison parameters taken into account included training time for the particular network and the average absolute output error in case of noisy input. For the case of the noisy input data, results for various levels of noise were studied.