A Feasible Solution Algorithm for Identifying Pareto-Based Multi objective Solutions For Subset Selection

• Activity Number: 358 - Contributed Poster Presentations: Section on Statistics in Epidemiology
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Joshua Lambert* and Greg Hawk and Katie Thompson and Arnold Stromberg
• Companies: University of Cincinnati and University of Kentucky and University of Kentucky and University of Kentucky

Abstract

The concept of pareto optimality has been utilized in the fields such as engineering, economics, and machine learning to understand fluid dynamics, consumer behavior, by identifying parameters that best optimize a set of m criteria. In model selection statisticians are often concerned with the model which has the single most optimal criterion (eg. AIC, R^2) before checking several other diagnostics. This strategy is multi-objective in nature but single objective in its numeric execution. This poster will first introduce the general framework of Pareto optimality and common strategies to attain and visualize its solutions. Next, a feasible solutions algorithm will be introduced as well as how the algorithm can be applied to the multi objective problem. Finally, simulation results for a fixed subset size bi-objective problem are discussed as well as an application to a Communities and Crime within United States dataset.