End-to-End Prediction of Parcel Delivery Time With Deep Learning for Smart-City Applications

The acquisition of massive data on parcel delivery motivates postal operators to foster the development predictive systems improve customer service. Predicting times successive being shipped out final depot, referred as last-mile prediction, deals with complicating factors such traffic, drivers’ behaviors, and weather. This work studies use deep learning for solving a real-world case last-mile time prediction. We present our solution under Internet-of-Things (IoT) paradigm discuss its feasibility cloud-based architecture smart city application. focus large-scale set provided by Canada Post, covering Greater Toronto Area (GTA). utilize an origin-destination (OD) formulation, in which routes are not available, but only start end points. investigate three categories convolutional-based neural networks assess their performances task. further demonstrate how modeling outperforms several baselines, from classical machine models referenced OD solutions. perform thorough error analysis across visualize features learned better understand model behavior, making interesting remarks predictability. Our provides end-to-end pipeline that leverages well weather accurately predict durations. believe system has potential user experience anticipation also aid logistics whole.