Data-Driven Optimization for Order Dispatching in Food Delivery Platforms: An Explainable Predict-then-Optimize Method

The advent of artificial intelligence has provided effective solutions to operational issues in on-demand food delivery platforms. When matching riders with orders, the platforms face the challenge of delivery delays. Optimizing order dispatch is a crucial approach to enhance delivery efficiency, with travel time prediction being central to the order dispatching problem. Utilizing a large-scale dataset released by a large food delivery platform in China, this paper proposes an explainable predict-then-optimize method that combines interpretability with improved optimization efficiency. By integrating explainable machine learning with optimization methods, it effectively reduces the number of delayed orders. Our results demonstrate that the proposed explainable predict-then-optimize method provides transparent and effective decision support for on-demand food delivery platform operators. Compared to traditional predict-then-optimize methods, our proposed approach reduces delay time by over 90% and shortens the average solution time by approximately 4.74 seconds.