Sensitivity analysis of an autonomous evolutionary algorithm

Abstract

When applying any heuristic, the user faces difficulties on deciding on the control parameters of the method. A generic sensitivity analysis to measure the interdependencies of the parameters from an autonomous evolutionary algorithm and their influence in the final result is shown. The Multi Dynamics Algorithm for Global Optimization is the base of the experiment. With only two parameters, it is a quasi-free parameter autonomous algorithm. The impact on the quality of the results on several multimodal standard problems applying different instances of those parameters has been studied. Excellent outcomes for sensitivity levels from 100 to 10-5 are found. The Logit model is used to determine the functioning parameters of the MAGO and for their mutual effects. Depending on the problem type, its dimensionality, and the expected precision, this work gives a priori configuring for the best performance of the MAGO. © Springer-Verlag Berlin Heidelberg 2012.