AI for Aid: Using ReliefNET-GNN to Enhance Transportation Resilience in Disaster Response

Abstract

Transport networks are essential for disaster response, yet identifying critical links within them remains a computationally intensive task, especially at scale. Traditional vulnerability assessment requires exhaustive simulations, limiting their practical use in time-sensitive crisis situations. We present ReliefNET-GNN, a graph neural network model that efficiently predicts the criticality of transport links relevant to humanitarian aid delivery. Trained on synthetic and real-world networks labeled with established vulnerability metrics, the model enables rapid inference without costly scenario evaluations.Preliminary results show that ReliefNET-GNN achieves comparable accuracy to traditional methods while drastically reducing computational time, making it suitable for real-time governmental decision-making during emergencies. This research contributes to the development of scalable, AI-driven tools that enhance disaster resilience and crisis management capabilities, particularly for public sector actors.