Easy Efficiency-Enhancement Technique for the ECGA

Several works have shown that population size affects the performance and efficiency of evolutionary optimization algorithms. There are works in the literature proposing techniques for the determination of the best size of a population using complex equations or synthetic benchmarks. Nevertheless, higher population size usually lead to more function evaluations and higher running time. On the other hand, lower population size in general implies a poor sampling of the search space and premature convergence of the algorithm. This paper investigates the influence of the first population on estimation of distribution algorithms. The basic idea is that the first population has a very larger effect on the estimation of distribution than the remaining populations. We analyzed the effects produced on an ECGA by different population sizes and distribution functions in the first population. We verified that the running population size can be smaller than the one estimated by the authors of the ECGA, if a larger initial population is used. Thus, it is possible the development of ECGAs requiring relatively lower running time and evaluations without decreasing the success rate. Therefore, the knowledge provided by this work can largely contribute for the improvement of general estimation of distribution algorithms.