Building fuzzy neural classifiers by fuzzy ID3 algorithm

This paper proposes a new method to building a fuzzy neural classifier which could be thought as a fuzzy system completed with a neural network. First we need to obtain an initial rule-based fuzzy classifier, then convert this system into a multilayer feedforward neural network architecture. Recently, automatically generating fuzzy rules from numerical data has received more and more attention. Based on thee traditional ID3 algorithm, we propose a fuzzy ID3 algorithm for feature space partitioning. The rule-based fuzzy classifier can be determined from the feature subspaces.