Graph Neural Networks for Spatio-Temporal Forecasting of Foot-and-Mouth Disease Risk Using Livestock Movement Traceability Data

Abstract

Foot-and-Mouth Disease (FMD) remains a critical threat to global livestock industries, causing severe economic losses and trade restrictions. This paper proposes a novel application of Temporal Graph Networks (TGNs) to forecast FMD outbreak risk with a four-week horizon using livestock movement traceability data. By modeling complex, time-evolving relationships between farms and geographic regions, the TGN framework captures the dynamic spatio-temporal dependencies that govern disease spread. We evaluate our model against several benchmarks, including Logistic Regression, LSTM, static Graph Convolutional Networks (GCNs), and a GCN-LSTM hybrid. Our results demonstrate that the proposed TGN model achieves superior performance, with an F1-score of 0.89 and an AUC-ROC of 0.94, significantly outperforming all baseline approaches. The study highlights the potential of advanced graph-based deep learning to enhance veterinary epidemiological surveillance and enable proactive disease control strategies through early warning systems.