Data-driven Arrival Time Suggestion for Drivers in Trucking Platform of Export Containers

Trucking platform has emerged in recent years for improving export freight logistics. As a bilateral platform, it needs to balance the satisfaction of exporters and truck drivers. A typical scenario is that given the required start loading time of the container at the exporter’s warehouse, the platform can make an arrival time suggestion to the assigned driver for reducing the expected time cost for both the exporter and the driver. This decision problem is nontrivial due to uncertainties in drivers’ punctuality and stay times at the warehouses as well as partial schedule and truck information possessed by the platform. Thus, we propose a data-driven arrival time suggestion model (ATSM) which computes the suggested arrival time in real-time by solving a stochastic program once a new pick-up task is assigned. Both feature matching- and residual-based strategies are proposed for constructing the predictive probability distributions. We prove properties of the single- and bi-objective formulations and design efficient algorithms for common families of punctuality distributions illustrated by numerical examples. Experiments on a real dataset show that the proposed ATSM can potentially reduce the average total time cost by 2.95∼7.58% against historical decisions over various parameter settings. Optimal decision patterns are affected by different task-, driver- and warehouse-specific features. Impact of key algorithm and model parameters are also discussed, leading to various managerial insights. Our single-objective stochastic program is a novel variant of the newsvendor problem whose theoretical properties and new implications to data-driven decision strategies are discussed, which may be of separate interest.