Contextual Optimizer Through Neighborhood Estimation for Prescriptive Analytics

We study black-box contextual optimization problems addressing challenges from a vast contextual space and heteroscedastic noise. To solve this problem, we derive an efficient sequential sampling rule with the awareness that observations acquired are used to infer conditional optimality. We first propose a consistent Shrinking Neighborhood Estimation (SNE) that estimates performance by the average performance in the neighborhood contexts. Then, we propose a Rate-Efficient Sampling rule that optimizes a lower bound performance of SNE-inferred contextual optimal decisions. We prove that the combined solution “Contextual Optimizer through Neighborhood Estimation” (CONE) leads to a stretched exponential decay rate of an upper bound of the decision loss. The methodology is deployed to address a staffing problem in a hospital call center. The result shows that the CONE method can deliver a staffing policy that makes a trade-off between current queue status and future arrival patterns.

History: Accepted by Bruno Tuffin, Area Editor for Simulation.

Funding: Financial support from the Centre for Next Generation Logistics; the Centre of Excellence in Modelling and Simulation for Next Generation Ports; and the National University Health System is gratefully acknowledged.

Supplemental Material: The software that supports the findings of this study is available within the paper and its Supplemental Information (https://pubsonline.informs.org/doi/suppl/10.1287/ijoc.2025.1134) as well as from the IJOC GitHub software repository (https://github.com/INFORMSJoC/2025.1134). The complete IJOC Software and Data Repository is available at https://informsjoc.github.io/.