Learning fuzzy concept prototypes using genetic algorithms

Most real world concepts are not precisely defined, and their boundaries are fuzzy. These concepts usually possess graded structures. Such concepts are called graded or fuzzy concepts. The prototype view was proposed for representing graded concepts. Another attempt at handling graded concepts is the work on fuzzy set theory introduced by Zadeh. This paper presents a genetic algorithmic approach to learning fuzzy prototypes. Given n attributes, a fuzzy prototype of a concept is a vector of n fuzzy membership functions, each for one attribute. In existing prototype learning systems, concept membership of an instance is determined using a distance measure. Using fuzzy prototypes, concept membership of an instance is determined by a collection of fuzzy membership functions. This approach has been implemented in a system named FuzzyProto.