Contextual Optimization Under Model Misspecification: A Tractable and Generalizable Approach

Contextual optimization problems arise in decision-making applications where historical data and contextual features are used to learn predictive models that guide optimal decisions. Practical applications often face model misspecification from incomplete knowledge of the data-generating process, leading to suboptimal decisions. Existing methods mainly address well-specified models, leaving a gap in the literature in handling misspecification. We propose a consistent, tractable and generalizable Integrated Learning and Optimization (ILO) framework that successfully addresses this gap in the literature.