Struggling with genetic algorithm parameter tuning? This paper addresses the challenge of reliably setting genetic algorithm parameters for consistent labeling problems, proposing a robust empirical framework based on factorial experiments. The approach begins with a graeco-latin square to study a wide range of parameter settings. Followed by fully crossed factorial experiments, allowing detailed analysis by logistic regression. A robust empirical framework permits an initial study of a wide range of parameter settings. The derived empirical models can be used to determine optimal algorithm parameters and shed light on interactions between parameters and their relative importance. Re-fined models are shown to be robust under extrapolation, offering valuable guidance for algorithm design and application.
Published in Evolutionary Computation, this paper aligns with the journal's focus on the theory and application of evolutionary algorithms. By providing an empirical framework for parameter tuning, the study contributes to the practical advancement of genetic algorithms in solving complex problems.