A fuzzy inference fitness function for evolutionary learning systems

Evolutionary learning algorithms generally employ genetic search methods and usually consist of a finite rcpctition of three steps at each generation: selection of thc parent chromosomes, recombination using crossovcr and mutation operations and a fitness function that describes the goodness individual members of each generation. The fitness ftinction is an iinportant component of the process, since it quantifies the performance of each individual in a generation and between generations. This paper prcsents a new inethod for selecting the fitness function for evolutionary learning. The approach is based upon fuzzy inference and employs the A-Law compander function in the expander mode. Results show that the approach provides better pcrforinance than classical fitness function mcthods.