On Partitioned Fitness-Distributions of Genetic Operators for Predicting GA Performance

Abstract

A statistical approach aimed at predicting the performance of a GA is presented. This approach is based on trying to mimic the fitness distribution of genetic operators. By modeling such fitness distributions, the effects of genetic operators can be simulated within the framework of virtual genetic algorithms (VGAs). An improved statistical model is provided for this purpose. Avery important issue is the partitioning of the data set used in the statistical analysis, so as to grasp better the structure of the real distribution.