A Framework for Studying the Effects of Dynamic Crossover, Mutation, and Population Sizing in Genetic Algorithms

We introduce a framework for controlling genetic algorithms and use it to study the effect of dynamically modulating genetic algorithm control. parameters on search behavior. Our framework includes techniques that can automatically design control strategies for genetic algorithms according to a given search performance metric. Many of the automatically designed strategies exhibit an exponentially decreasing mutation rate behavior. We present experimental results indicating that exponentially decreasing the mutation rate over time contributes more towards an increase on online and offline search performance than populations size or crossover rate modulation.