Learning fuzzy rules from artificial neural nets

Abstract

An algorithm is given for extracting fuzzy rules from a neural net model called a self-organizing feature map. These rules can also be transformed into a linguistic form. The algorithm gives an interpretation of the map after the learning process by describing its end configuration with fuzzy rules. This approach can be used in the area of knowledge acquisition if only a vast set of unclassified data of a given domain is available. The underlying ideas of the knowledge extraction algorithm are presented. The generation of membership functions is depicted. The process of creating rules out of these membership functions is described. The results of testing the algorithm with some real data sets are presented.<>