Capacity planning in e-commerce logistics using a hybrid machine learning model

Official URL: https://dx.doi.org/10.1109/ASYU58738.2023.10296626

Due to the increase in e-commerce demand, the escalating exponential growth of congestion within transportation systems has reached a critical juncture, significantly impinging upon the punctual delivery of routine parcels and groceries. A crucial challenge to be resolved is that drivers operate under time constraints within a specific number of deliveries and have a restricted daily capacity that requires more comprehensive and effective capacity planning. In this paper, an integrated approach composed of clustering and stages of regressions is developed in which the delivery information of the cross-docks in a cluster is utilized to predict the daily delivery capacity of a fleet that starts its routes from a cross-dock depot in a specific time slot. Each prediction specifies the amount of delivery in total and for a given cross-dock, within a certain time slot of the day by the drivers. Our results show that for most of the clusters, the proposed GPR model outperforms other state-of-the-art regression methods. Also, the model is daily updated using data from shipments delivered on the same day. This ensures adaptability to unforeseen events and factors like special occasions (e.g., Black Friday or Christmas) in the logistics domain.