The extraction of Sugeno fuzzy rules from neural networks

摘要

This paper describes an algorithm for the extraction of Sugeno fuzzy rules from feedforward neural networks without a dedicated architecture. Three application examples of the algorithm are presented: exclusive or, Iris classification and prediction of chemical properties. The algorithm exposes the strategy used by the network to solve the problem and provides insight into the linearities in the data set. The number of rules generated by the algorithm is O (number of patterns).