Figure 4
Benchmark of various combinations of mutation and crossover operators. The six box plots depict the dependency of fitness on a multitude of combinations of mutation and crossover operators available for binGA (a-c) and ES (d-f). All plots are limited to a fitness of 60. The more detailed the model the better the overall fitness that can be obtained. For binGA almost all combinations of mutation and crossover operators yield good fitness values. The only two exceptions are no mutation with one- or n-point crossover, which shows the worst fitness on all three models. The ES is more sensitive to alternative combinations of mutation and crossover operators. The abbreviations cma, correlat., no mutat., and 1/5 success stand for covariance matrix adaptation, correlated, and no mutation and for the 1/5th success rule. The elaborated comparison shows that some settings of the ES lead to equally good or even better results than that of binGA. However, most ES-settings cannot compete with the fitness values found for most binGA settings.