A visual tool for analyzing the behavior of metaheuristic algorithms

In this paper a tool for supporting visual analysis of the behavior of metaheuristic algorithms focused to solve the Bin Packing Problem is proposed. Traditionally, metaheuristics have been analyzed monolithically, by means of solving a set of input instances and analyzing the output solutions. However, due to the stochastic features of metaheuristics, for a given instance and several executions of a metaheuristic, solutions may be excellent as the optimal solution or can be bad as the worst solution, which makes difficult to understand the behavior of the metaheuristic algorithm. In this sense, the proposed tool allows storing and reproducing the algorithm behavior during any execution making possible the metaheuristic analysis and improvement. To validate the tool the Hybrid Grouping Genetic Algorithm for Bin Packing (HGGA_BP) was analyzed while it was solving the WAE_GAU1 benchmark of instances. Based on the analysis results some improvements to the Metaheuristic algorithm were identified. The improvements allowed increasing the metaheuristic efficiency by 48%.