Predicting Emerging Supply Chain Bottlenecks with Graph Neural Networks

Modern supply chains constitute complex, dynamic networks with intricate interdependencies across multiple tiers. Traditional analytical methods often struggle to capture the network effects and cascading impacts inherent in these systems, limiting their ability to predict emerging bottlenecks and systemic vulnerabilities accurately. This article explores the application of Graph Neural Networks (GNNs) for modeling multi-tier supply chain structures and predicting disruption propagation. By representing suppliers, manufacturers, distributors, and logistics hubs as nodes and their relationships as edges, GNNs can learn complex relational patterns and network dynamics. The proposed temporal hybrid graph attention model not only leverages diverse data sources to identify hidden vulnerabilities but also extends established theoretical frameworks in supply chain management, including Resource Dependence Theory, Complex Adaptive Systems, and Transaction Cost Economics. Our approach operationalizes abstract theoretical constructs through computational mechanisms, providing empirical validation of key theoretical predictions while enabling dynamic extensions of traditionally static theories. Results based on comprehensive empirical data spanning 4,327 companies across 23 countries and 7 industries demonstrate the potential of GNNs to provide earlier warnings and more precise localization of emerging systemic risks, enabling proactive mitigation strategies before disruptions cascade throughout the supply chain. Furthermore, we address a fundamental methodological challenge in predictive systems—the intervention paradox—by developing an evaluation framework that accounts for the impact of successful preventive actions on model assessment, establishing a foundation for future research in intervention-aware prediction across multiple domains.