Performance Classification of Genetic Algorithms on Continuous Optimization Problems

摘要

Modelling the behaviour of algorithms is the realm of Evolutionary Algorithm theory. From a practitioner's point of view, theory must provide some guidelines regarding which algorithm/parameters to use in order to solve a particular problem. Unfortunately, most theoretical models of evolutionary algorithms are difficult to apply to realistic situations. Recently, there have been works that addressed this problem by proposing models of performance of different Genetic Programming Systems. In this work, we complement previous approaches by proposing a scheme capable of classifying the hardness of optimization problems based on different difficulty measures such as Negative Slope Coefficient, Fitness Distance Correlation, Neutrality, Ruggedness, Basins of Attraction, and Epistasis. The results indicate that this procedure is able to accurately classify the performance of the GA over a set of benchmark problems.