On fitness distance distributions and correlations , GA performance , and population size of fitness functions with translated optima

Fitness distance correlation has been proposed as a n nalysis tool for the fitness function landscape of genetic algorithms. It has been shown that it wo rks in some cases, but there are also counterexamples. We studied the fitness distance correlati on and the optimal population size by two fitness functions with a translational minimum and with a m eta-genetic algorithm. We discovered that the fitness distance distribution and the actual optimi zation performance depend strongly on the place of the optimum. We also found that fitness distance co rrelation explains moderately the variations in the performance of the genetic algorithm and also s ome of the variations in the optimal population size obtained by a meta-genetic algorithm.