A genetic-based method for learning the parameters of a fuzzy inference system

Abstract

Fuzzy inference systems (FIS) provide models for approximating continuous, real valued functions. The successful application of fuzzy reasoning models depends on a number of parameters, such as the fuzzy partition of the input/output universes of discourse, that are usually decided in a subjective manner (traditionally, fuzzy rule bases are constructed by knowledge acquisition from human experts). This paper presents a flexible genetic based method for learning the parameters of a FIS from examples such as the subjectivity not to be involved at all. We show that applying this method it is possible to obtain better performances for the FIS or, for the same performances, a less complex structure for the system.