<![CDATA[Supply chain disruptions in regional food distribution networks can generate cascading impacts due to complex interdependencies among producers, warehouses, distributors, retailers, and other supply chain entities. Existing risk prediction approaches often rely on static representations and independent feature analysis, limiting their ability to capture dynamic network evolution, risk propagation mechanisms, and interpretable decision support. This study proposes a Resilient Supply Chain Risk Prediction framework using graph-based learning for a regional rice distribution network in Aceh Besar Regency and Banda Aceh City, Indonesia. The proposed framework integrates Dynamic Supply Chain Graphs, Temporal Graph Neural Networks (Temporal GNNs), risk propagation analysis, and explainable artificial intelligence (XAI) to model evolving supply chain relationships and identify disruption vulnerabilities. Operational data from rice mills, warehouses, distributors, and traditional markets are transformed into sequential dynamic graph representations to capture changes in supply availability, inventory conditions, transportation relationships, demand variations, and disruption patterns over time. Experimental results demonstrate that the proposed Temporal GNN outperforms Random Forest, XGBoost, and LSTM models, achieving 92.40% accuracy, 92.49% F1-score, and an AUC-ROC of 0.968. Furthermore, risk propagation analysis identifies critical nodes and disruption pathways within the supply chain network, while explainability analysis using GNNExplainer and SHAP reveals that lead time, inventory level, and supplier reliability are key factors influencing disruption risks. The proposed framework also supports resilience-oriented decision-making by enabling proactive mitigation strategies, including route adjustment, inventory redistribution, and prioritization of critical supply chain entities. This study contributes to supply chain risk analytics by providing an interpretable dynamic graph-based learning framework that simultaneously captures network evolution, temporal risk patterns, cascading disruption behavior, and resilience improvement strategies for regional food distribution systems]]>
Copyrights © 2026
