A fuzzy-GRNN classifier for pattern recognition

We present a scheme for combining fuzzy c-means clustering with generalized regression neural network for improving its pattern classification efficiency. The membership grades of data points produced by fuzzy c-means clustering have been used to define weights for data points in the feature space according a logarithmic measure of uncertainty similar to the Shannon's entropy. The method improves classification results from 1% to 17% for 9 data sets analyzed here.