Simulation screening experiments using lasso-optimal supersaturated design and analysis: a maritime operations application

Abstract

Screening methods are beneficial for studies involving simulations that have a large number of variables where a relatively small (but unknown) subset is important. In this paper, we show how a newly proposed Lasso-optimal screening design and analysis method can be useful for efficiently conducting simulation screening experiments. Our approach uses new criteria for generating supersaturated designs, and a new algorithm for selecting the optimal tuning parameters for Lasso model selection. We generate a 24x69 Lasso optimal supersaturated design, illustrate its potential with a numerical evaluation, and apply it to an agent-based simulation of maritime escort operations in the Strait of Gibraltar. This application is part of a larger project that seeks to leverage simulation models during the ship design process, and so construct ships that are both cost effective and operationally effective. The supersaturated screening design has already proved beneficial for model verification and validation.

Screening methods are beneficial for studies involving simulations that have a large number of variables where a relatively small (but unknown) subset is important. In this paper, we show how a newly proposed Lasso-optimal screening design and analysis method can be useful for efficiently conducting simulation screening experiments. Our approach uses new criteria for generating supersaturated designs, and a new algorithm for selecting the optimal tuning parameters for Lasso model selection. We generate a 24x69 Lasso optimal supersaturated design, illustrate its potential with a numerical evaluation, and apply it to an agent-based simulation of maritime escort operations in the Strait of Gibraltar. This application is part of a larger project that seeks to leverage simulation models during the ship design process, and so construct ships that are both cost effective and operationally effective. The supersaturated screening design has already proved beneficial for model verification and validation.