Data-driven imitation learning-based approach for order size determination in supply chains

Past studies have attempted to formulate the order decision-making behaviour of humans for inventory replenishment in dynamic stock management environments. This paper investigates whether a data-driven approach like machine learning can imitate the order size decisions of humans and consequently enhance supply chain performances. Accordingly, this paper proposes a supervised machine learning-based order size determination approach. The proposed approach is initially executed using the order decision data collected from a simulated stock management environment similar to the 'beer game'. Subsequent comparative analysis shows that the proposed approach successfully enhances all supply chain performance measures compared to other well-known ordering methods. Additionally, the proposed approach is validated on a retail case study to investigate its efficacy. This paper thus focuses on extending the past works reported in the literature by modelling human order decision-making as data-driven imitation learning and contributing to machine learning applications for order management. [Submitted: 19 August 2021; Accepted: 16 February 2022]