Determining Relative Importance and Best Settings for Genetic Algorithm Control Parameters
Kevin L. Mills, James J. Filliben, Andrea Haines
We intend to use a classic genetic algorithm (GA) to steer a population of cloud-computing simulators into behavioral directions that reveal degraded performance and system collapse. Such a method could serve as a design tool, empowering system engineers to identify and mitigate low-probability, costly failure scenarios. In the existing GA literature, we uncovered conflicting opinions and evidence regarding key GA control parameters, and the best settings to adopt. Consequently, we designed and executed an experiment to determine the relative importance and best settings for seven GA control parameters, when applied across a set of numeric optimization problems drawn from the literature. This paper describes our experiment design, analysis methods and results. We found that crossover and mutation most significantly influence GA success, followed by population size and reboot point, while elite selection and selection method ranked third. Precision used within the chromosome to represent numerical values had least influence. This paper makes two main contributions: (1) we define an experiment design and analysis approach that can be adapted to determine relative importance and best settings for control parameters in any evolutionary computation algorithm and (2) for a classic GA we determine the relative importance and best settings for seven control parameters. Our findings are robust over 60 numeric optimization problems.