A Simple Method to Extract Fuzzy Rules by Measure of Fuzziness

Abstract

To easily construct an efficient rule-based system for pattern classification problem is an important research topic. In this paper, we use a simple method to build up the fuzzy rules directly from numerical input-output data. First, we extract fuzzy rules from different class regions which are confined by training data. These confined regions construct the basic element-nodes in the second layer for our proposed neural network structure. When these confined regions are overlapped, a recursive process is applied possibly to set up additive fuzzy rules in these uncertainty-overlap regions. The stop criterion is referred in a measure of fuzziness to keep creating the efficient rules for the recursive process. All additive fuzzy rules are added into this fuzzy-neural network. Finally, the method is compared with other algorithms using the Fisher iris data and a set of pseudo iris data for performance evaluation.