Budget constrained model selection for multiple linear regression

Abstract

We introduce a multiple linear regression model methodology for selecting best subsets of independent variables subject to active budget constraints. Computational studies on multiple data sets to evaluate the performance of our methods find them to be competitive and robust compared with traditional model selection methods, and practical while providing the added benefit of model selection subject to a constrained budget for variables.

Keywords:

• Analytics
• Multiple linear regression
• Variable selection
• Budget
• MINLP