Post-hazard supply chain disruption: Predicting firm-level sales using graph neural network

Abstract

Evaluating the damage caused by supply chain disruptions triggered by natural hazards has an important role in supply chain risk management. This study proposes a novel approach to predict annual sales change rates for firms post-hazard. The approach employs graph neural networks, a machine learning technique that considers both internal factors such as sales and number of employees and external factors such as inter-firm relations and damages incurred from hazards. The model was trained on firm and flood data from previous floods in Japan and was found to outperform baseline models that did not consider inter-firm relations. To address the model's explainability problem, we employed Explainable Artificial Intelligence techniques to identify the factors influencing sales post-disruption and analysed the importance of firms and trades in the supply chain from a spatial perspective. The findings highlight the crucial role of business partners in a firm's supply chain and the impact of distance between suppliers on a firm's sales. The study contributes to improved supply chain risk management practices in industry as well as government, ultimately enhancing supply chain resilience.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 21 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.