Skip to content
Book Open access

Graph Neural Delay Differential Equations for Continuous Epidemic Forecasting

Nov 2026 · Proceedings of the 35th ACM International Conference on Information and Knowledge Management · 0 citations · 51 references

Abstract

Reliable epidemic forecasting is essential for early warning and proactive public health decision-making. Existing spatio-temporal forecasting methods often model regional interactions on fixed discrete grids and rely on instantaneous or uniformly delayed transmission assumptions, limiting their ability to capture asynchronous epidemic spread and irregular surveillance records. To address these limitations, we propose EpiDDE, a continuous-time delay-aware framework based on graph neural delay differential equations for epidemic forecasting. EpiDDE first estimates latent epidemic delay priors from surveillance sequences, including both within-region temporal delays and cross-region propagation delays. It then constructs a Delay-Aware Transmission Graph (DATG) by combining sequence-derived delay priors with learnable regional representations, enabling data-driven modeling of directed and heterogeneous delayed dependencies. Conditioned on the learned DATG, EpiDDE evolves regional epidemic states through a graph-coupled neural DDE driven by delayed self-region and cross-region messages. Extensive experiments on four real-world epidemic datasets show that EpiDDE achieves competitive epidemic forecasting performance across multiple prediction horizons. Further analyses demonstrate the effectiveness of delay estimation, DATG learning, and DDE-based continuous evolution, particularly for long-horizon forecasting and irregular observations.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.