Technical Note—Data-Driven Sample Average Approximation with Covariate Information

Using Side Information to Improve Decision Making Under Uncertainty In many real-world planning settings, side information (also called covariate information, contextual information, or features) can be used to improve the estimates of uncertain parameters. Over the past decade, there has been growing interest in data-driven approaches to stochastic programming that take advantage of such side information. In “Data-Driven Sample Average Approximation with Covariate Information,” Kannan, Bayraksan, and Luedtke investigate two flexible data-driven frameworks that integrate a machine learning prediction model within a sample average approximation (SAA) of a stochastic programming problem, including a novel framework that leverages leave-one-out residuals for scenario generation. They establish conditions on the data generation process, the prediction model, and the stochastic program under which the solutions of these data-driven contextual SAAs exhibit asymptotic and finite sample convergence guarantees. Furthermore, they provide examples illustrating that these data-driven formulations can outperform existing methods in the limited data regime, even if the prediction model is misspecified.