A hybrid genetic fuzzy neural network algorithm designed for classification problems involving several groups

We propose a multigroup classification algorithm based on a hybrid genetic fuzzy neural net (GFNN) framework. Recent results on evolutionary computati on and fuzzy neural network methodology are combined to effectively adapt t he membership functions of the fuzzifier and the defuzzifier to the data se t. Separate membership functions are defined for each dimension in the fuzz ifier and for each fuzzy output group in the defuzzifier. The signal inhere nt in the fuzzifier is aggregated by a suitable T-norm and transmitted to t he defuzzifier. The defuzzifier aggregates the response, i.e., the predicte d group membership, by a suitable conorm. If misclassifications occur durin g training, the membership functions of both the fuzzifier and the defuzzif ier are adapted by a systematic, robust procedure. The algorithm is success fully tested with real economic data. In total, the GFNN performs as good a s the best of the competing methods in our test. The results suggest econom ically meaningful interpretations. (C) 2000 Elsevier Science B.V. All right s reserved.