Classification of fuzzy input patterns by neural networks

In this paper we propose an approach to the classification of fuzzy input patterns by a multilayer feedforward neural network. Our neural network can handle linguistic inputs such us “small”, “medium” and “large” as well as fuzzy numbers such as “about 2” and “approximately 3”. First we briefly describe the input-output relation of our neural network for fuzzy input patterns. A fuzzy input pattern is mapped to fuzzy number outputs. Next we propose a classification method of the fuzzy input pattern. In the proposed method the grade that the fuzzy input pattern belongs to each class is calculated in the framework of possibility theory. Because our approach can handle linguistic values as inputs, it can also be utilized as a fuzzy rule generation method from the trained neural network