Multi-commodity, multi-echelon stacked machine learning models for replicating global supply chain structures with limited, publicly available information

Purpose The ability to replicate global supply chain (SC) structures is important for maintaining and improving the operations of global SCs. This structure, while known to specific manufacturers and key suppliers, is not generally published and may not be totally transparent to all stakeholders. This paper presents a methodology for replicating the structure of global SCs with limited, publicly available data. Design/methodology/approach The proposed SC replication method decomposes the multi-echelon SC into a set of duo-echelon models, each focused on a single commodity, which are reconstituted for a complete representation of the SC for improved computational efficiency. Flows of raw materials, middle products (product components) and end-products for each duo-echelon model are obtained from a developed stacked machine learning method using only publicly available information. The duo-echelon models are integrated into a final multi-echelon, multi-commodity SC representation. Findings The method was applied to a case study of lithium-ion battery production. The findings underscore the model’s capability to capture complex trade relationships and global SC dynamics. Further, the method outperformed numerous other approaches. Originality/value This paper proposes a novel approach to estimating global SC structures, including complexities from their multi-commodity and multi-echelon nature, using publicly available data.