Bilevel optimization for feature selection in the data-driven newsvendor problem

Breno Serrano, Stefan Minner, Maximilian Schiffer, Thibaut Vidal

Abstract: Highlights•We introduce a bilevel program for feature selection for the data-driven newsvendor.•We select a subset of features, minimizing the newsvendor cost on a validation set.•We reformulate the bilevel program into a single-level mixed integer linear program.•Our bilevel program outperforms regularization-based methods in feature recovery.•In most cases, we observe an improvement in out-of-sample cost performance.