Understanding Food-Delivery Lateness Through Spatiotemporal Analysis and Machine Learning

Meituan is a China-based technology company founded in 2010 that operates a food delivery service handling thousands of orders every day. The estimated delivery time for these orders is sometimes wrong. This paper will attempt to determine the cause of order lateness. The data used is from the Transportation Science and Logistics Data-Driven Research Challenge, in which participants are tasked with developing data-driven optimization models using data provided by Meituan. The data covers one anonymous city over one week, with 654,343 data points focused on delivery. Using this data, we identify patterns, such as the temporal difference between prebooked orders, and discover rush hours and hotspots with a higher percentage of orders, using methods like H3 clustering. Afterward, using machine learning from three different models, logistic regression, random forest, and XGBoost, and evaluating the ROC-AUC score as a measure of accuracy, we find feature importance to determine the most important features affecting the lateness of an order. This paper can address inaccurate initial estimated delivery time, increase customer satisfaction, and improve delivery process efficiency.