Route Prediction for Instant Delivery

Abstract

Instant delivery has drawn much attention recently, as it greatly facilitates people's daily lives. Unlike postal services, instant delivery imposes a strict deadline on couriers after a customer places an order online. Therefore it is critical to dispatch the order to an appropriate courier to guarantee the timely delivery. Ideally couriers should choose the optimal routes with the lowest overdue rate (i.e., the rate of the deliveries that are not finished in time) and the minimal distance. In practice, however, decision-making of the couriers is quite complex because individuals have different psychological perception of the environments (e.g., distance) and delivery requirements (e.g., deadline). To well predict their behaviors, we design multiple features to model the decision-making psychology of individual couriers and predict couriers' route with a machine learning algorithm. In particular, we reveal that perceived distance is the main factor influencing couriers' decision, which should be modeled based on the subjective understanding of the actual distances. Our design is implemented, deployed and evaluated on Ele.me, which is one of the largest instant delivery platforms in the world. Experimental results show that the overdue rate can be reduced by 48.02%, which is a significant improvement.