A Support Vector Machine-Inspired Evolutionary Approach for Parameter Tuning in Metaheuristics (Working Paper)

Summary. The paper presents a novel, combined methodology to target parameter tuning. It uses Latin hypercube sampling to generate a diverse, large set of config-urations for the variables to be set. These serve as input for the metaheuristic to be tuned and an extensive data set, with the parameter values and the success rate obtained by the algorithm, is formed. The collection is next subject to regression by means of a recent evolutionary engine for support vector machine learning. The investigations on tuning an evolutionary algorithm for function optimization led to interesting insights on a simple, unconstrained evolution of the structure and coeffi-cients of the underlying regression function. The approach can be further improved in prediction accuracy, while also enhanced to target multiobjectivity and discovery of the best set of performing parameters for the metaheuristic to be tuned.