Abordagem co-evolutiva hierárquica para geração automática de sistemas nebulosos

Abstract

This work focuses on the problem of automatic generation of fuzzy systems through evolutionary computation, specifically using the approach of co-evolution. Coevolution is based on the idea of modular modeling of the problem subcomponents. In this work the subcomponents are represented by different species, which have a collaborative relation among them. The fuzzy system to be created has the objective of pattern classification. Basically, the evolutionary scheme is composed by four different species, which have a hierarchical collaboration both in the generation of the species and in the fitness determination of the individuals of these species. These species are organized in levels, where the contribution in the species generation happens from the lowest to highest levels and the contribution in the fitness determination happens from the highest to lowest levels. The results obtained indicate that the studied approach is very promising and, through its use, one can generate efficient classification systems, that present a performance similar to other approaches found in the literature.