Generating Exact Optimal Designs via Particle Swarm Optimization: Assessing Efficacy and Efficiency via Case Study

TL;DR

This paper evaluates particle swarm optimization (PSO) for generating optimal experimental designs, demonstrating its efficiency and effectiveness compared to traditional local optimizers through extensive case studies.

Contribution

It introduces a preferred PSO variant for design generation and provides empirical evidence of its superior efficiency and efficacy over existing methods.

Findings

PSO can generate high-quality designs with high probability in a single run.

The preferred PSO variant outperforms local optimizers like coordinate exchange.

PSO offers a cost-effective alternative for practitioners in experimental design.

Abstract

In this study we address existing deficiencies in the literature on applications of Particle Swarm Optimization to generate optimal designs. We present the results of a large computer study in which we bench-mark both efficiency and efficacy of PSO to generate high quality candidate designs for small-exact response surface scenarios commonly encountered by industrial practitioners. A preferred version of PSO is demonstrated and recommended. Further, in contrast to popular local optimizers such as the coordinate exchange, PSO is demonstrated to, even in a single run, generate highly efficient designs with large probability at small computing cost. Therefore, it appears beneficial for more practitioners to adopt and use PSO as tool for generating candidate experimental designs.