Investigating the Effect of Hyperparameter Values and Size on Swarm Optimization Effectiveness
George Tambouratzis · 2022 IEEE Symposium Series on Computational Intelligence (SSCI) · 2022
The present manuscript investigates the correlations between the control hyperparameters of a particle swarm optimization algorithm, the size of the swarm, and the algorithm's effectiveness in locating optimal solutions. This is achieved by analyzing a large volume of experiments that cover a range of swarm sizes using a selection of twenty-one standard benchmark functions. The findings are further analyzed via a ranking process to determine the swarm behavior. It is found that swarms below a specific minimum size do not achieve a good optimization solution. In addition, specific hyperparameter combinations consistently achieve very good solutions in optimization tasks, while other combinations consistently demonstrate a poor optimization behavior. Statistical tests are used to analyze the results comprehensively.