Scalability of multi-objective evolutionary algorithms for solving real-world complex optimization problems

Abstract

The use Multi-Objective Evolutionary Algorithms (MOEAs) to solve real-world multi-objective optimization problems often finds a problem designated by the curse of dimensionality. This is mainly because the progression of the algorithm along successive generations is based on non-dominance relations that practically do not exist when the number of objectives is high. Also, the existence of many objectives makes the choice of a solution to the problem under study very difficult. Several methods have been proposed in the literature to reduce the number of objectives to use during the optimization process. In the present work, a methodology to reduce the number of objectives is proposed. This method is based on DAMICORE (DAta MIning of COde REpositories), a machine-learning algorithm proposed by the authors. A theoretical comparison with a similar machine learning approach is made, pointing out some advantages of using the proposed algorithm using a benchmark problem designated by DTLZ5. Also, a real problem is used to show the effectiveness of the methodology.