Extraction of Fuzzy Rules Using Fuzzy Neural Networks with Forgetting
Motohide Umano, Shiro Fukunaka, Itsuo Hatono, Hiroyuki Tamura · Transactions of the Society of Instrument and Control Engineers · 1996
We extract fuzzy rules from data using a fuzzy neural network without a prior information. First, we set the number of initial fuzzy rules based on the number of training data. Next, we generate the specified number of initial fuzzy rules that have less wasteful membership functions using self-organization learning by T. Kohonen. Finally, we tune and prune fuzzy rules using the similar method to back-propagation learning with forgetting by M. Ishikawa. We apply the method to problems of the iris classification by R.A. Fisher and the diagnosis of electric transformer by gas in oil, and we have very good results.