Multiobjective Evolutionary Component Effect on Algorithm behavior

The performance of multiobjective evolutionary algorithms (MOEAs) varies across problems, making it hard to develop new or apply existing ones problems. To simplify the development and application algorithms, there has been an increasing interest in their automatic design from components. These automatically designed metaheuristics can outperform human-developed counterparts. However, is still unknown what are most influential components that lead improvements. This study specifies a methodology investigate effects final configuration algorithm. We this tuned Multiobjective Evolutionary Algorithm based on Decomposition (MOEA/D) by iterated racing (irace) package constrained problems 3 groups: (1) analytical real-world (2) artificial (3) simulated real-world. then compare impact algorithm terms Search Trajectory Networks (STNs), diversity population, anytime hypervolume values. Looking at objective space behavior, MOEAs studied converged before half search generally good HV values For improving end run. In decision we see diverse set trajectories STNs more similar frequently reach optimal solutions other