Graph Neural Network-Based Predictive Modeling for Enhanced Supply Chain Resilience against Multi-Modal Disruptions

Abstract

Global supply chains have become increasingly vulnerable to multi-modal disruptions, ranging from natural disasters and pandemics to geopolitical conflicts and cyber-attacks. Traditional approaches to resilience modeling often fail to capture the complex interdependencies that exist within modern supply networks. This paper introduces a novel Graph Neural Network (GNN) based framework for predictive modeling of supply chain resilience that leverages the inherent network structure of supply systems. Specifically, we employ Graph Attention Networks (GATs) with multi-head attention mechanisms to identify critical vulnerabilities and predict node-level resilience against various disruption scenarios. Our approach achieves 93.33% accuracy and 0.9630 F1 score in resilience classification tasks, outperform- ing traditional machine learning methods. Through attention mechanism analysis, we identify key structural dependencies and vulnerability patterns that can inform targeted resilience strategies. Our framework provides supply chain practitioners with an interpretable, data-driven approach to disruption risk management, enabling proactive rather than reactive resilience planning in complex global supply networks.