Integrating prediction with optimization : models and applications in transportation management

Abstract

Prediction and optimization are the foundation of many real-world analytics problems in various disciplines. As both can be challenging, they are usually treated sequentially in existing studies, where the prediction problem is dealt with in the first stage, followed by the optimization problem in the second stage, which is called the predict-then-optimize paradigm. Specifically, the unknown parameters in the optimization problem are first predicted by the prediction model and are then input to the optimization model to generate the optimal decisions. However, prediction models in the first stage are intended to minimize the prediction error, while ignoring the structure and property of the downstream optimization problem and how the predictions will be used. Consequently, suboptimal decisions might be generated. This editorial piece discusses current popular frameworks to integrate prediction with optimization, namely the smart “predict, then optimize” framework and the predictive prescription framework with examples in the transportation area provided. The article ends with proposing several promising research directions for future research.