Possibilistic Rules from Fuzzy Prototypes

A new interpretation of description labels in fuzzy rules is introduced based on the idea of fuzzy prototypes. For each label L i in a description set we identify a fuzzy prototype resulting in a possibility distribution quantifying the possibility that L i is the most appropriate label to describe a given input value. A rule induction algorithm is then proposed for learning Takagi-Sugeno style rules from data, within this label representation framework. Given an appropriate inference and defuzzification method we then demonstrate the potential of this approach by its application to a number of benchmark regression problems.