Delivery Time Prediction Based on Multi-Task Learning

• 1、School of Computer and Information Technology, Beijing JiaoTong University, Beijing 100044
• 2、Beijing Key Lab of Traffic Data Analysis and Mining, Beijing 100044

Abstract：Delivery time prediction ( i.e., to predict the arrival time of package delivery at any time) is one of the most important service in the logistics field. On the one hand, accurate prediction of delivery time can provide customer more punctual service and alleviate customer\'s waiting anxiety. On the other hand, predicting the delivery time is beneficial to the route planning for couriers, thus the delivery efficiency can be improved. However, in the real scenarios, multiple factors, dynamics and multiple destinations bring huge challenges to the package delivery time prediction. In this paper, a Multi-Task model for Delivery Time (MTDTN) prediction is proposed, to learn to estimate the arrival time of package delivery from couriers\' massive historical spatiotemporal trajectories. MTDTN takes full advantages of external factors that may affect the delivery time, and it utilizes geographic information encoder, convolution operations and Bi-LSTM to capture the spatiotemporal information in the trajectories. Moreover, for the first time, we use the multi-task learning to simultaneously predicting the delivery time as well as the delivery sequence. The model performance is enhanced by introducing the delivery order prediction as an auxiliary task. Extensive experiments on a real-world dataset demonstrate the superiority of MTDTN.

Keywords： Delivery time prediction Spatial-temporal trajectories Bi-LSTM Multi-Tasklearning