BACKPROPAGATION LEARNING ALGORITHMS FOR CLASSIFICATION WITH FUZZY MEAN-SQUARE ERROR

Most of the real life classification problems have ill defined, imprec ise or fuzzy class boundaries. Feedforward neural networks with conven tional backpropagation learning algorithm are not tailored to this kin d of classification problem. Hence, in this paper, feedforward neural networks, that use backpropagation learning algorithm with fuzzy objec tive functions, are investigated. A learning algorithm is proposed tha t minimizes an error term, which reflects the fuzzy classification fro m the point of view of possibilistic approach. Since the proposed algo rithm has possibilistic classification ability, it can encompass diffe rent backpropagation learning algorithm based on crisp and constrained fuzzy classification. The efficacy of the proposed scheme is demonstr ated on a vowel classification problem. (C) 1998 Elsevier Science B.V.