Abstract

Decision-focused learning(DFL) is an emerging paradigm that integrates machinelearning (ML) and constrained optimization to enhance decision quality by training MLmodels in an end-to-end system. This approach shows significant potential to revolutionizecombinatorial decision-making in real-world applications that operate under uncertainty,where estimating unknown parameters within decision models is a major challenge. Thispaper presents a comprehensive review of DFL, providing an in-depth analysis of bothgradient-based and gradient-free techniques used to combine ML and constrained opti-mization. It evaluates the strengths and limitations of these techniques and includes anextensive empirical evaluation of eleven methods across seven problems. The survey alsooffers insights into recent advancements and future research directions in DFL