MSGNN: Multi-scale Spatio-temporal Graph Neural Network for Epidemic Forecasting

TL;DR

MSGNN tackles epidemic forecasting by modeling multi-scale spatio-temporal dynamics. It introduces a graph learning module to capture long-range trans-regional connectivity ($A_s$) and short-range signals ($A_c$), and a multi-scale fusion-based graph convolution to extract scale-specific and scale-shared patterns across micro- and macro-scale graphs $G=(V,E)$ with $V=V_c\cup V_s$ and $E=E_c\cup E_s$. The forecasting pipeline yields $\hat{Y}$ via $\hat{Y}=\theta_f H_{out}$, with an emphasis on accuracy, robustness, and interpretability demonstrated on US COVID-19 data. Overall, MSGNN outperforms state-of-the-art baselines, offering actionable insights and reliable forecasts for epidemic management.

Abstract

Infectious disease forecasting has been a key focus and proved to be crucial in controlling epidemic. A recent trend is to develop forecast-ing models based on graph neural networks (GNNs). However, existing GNN-based methods suffer from two key limitations: (1) Current models broaden receptive fields by scaling the depth of GNNs, which is insuffi-cient to preserve the semantics of long-range connectivity between distant but epidemic related areas. (2) Previous approaches model epidemics within single spatial scale, while ignoring the multi-scale epidemic pat-terns derived from different scales. To address these deficiencies, we devise the Multi-scale Spatio-temporal Graph Neural Network (MSGNN) based on an innovative multi-scale view. To be specific, in the proposed MSGNN model, we first devise a novel graph learning module, which directly captures long-range connectivity from trans-regional epidemic signals and integrates them into a multi-scale graph. Based on the learned multi-scale graph, we utilize a newly designed graph convolution module to exploit multi-scale epidemic patterns. This module allows us to facilitate multi-scale epidemic modeling by mining both scale-shared and scale-specific pat-terns. Experimental results on forecasting new cases of COVID-19 in United State demonstrate the superiority of our method over state-of-arts. Further analyses and visualization also show that MSGNN offers not only accurate, but also robust and interpretable forecasting result.

Figures (5)

• Figure 1: A Comparison between single-scale based and multi-scale based epidemic spatio-temporal graph. The additional macro scale helps model long-range connectivity, which significantly reduces the hops of epidemic transmission path between distant areas.
• Figure 2: Pipeline of the proposed MSGNN. The pipeline consists of two main components, i.e. the graph learning module and the multi-scale graph convolution module.
• Figure 3: The per-week evaluation results for model robustness experiments, all the results are presented as box plots.
• Figure 4: Case studies for four counties with the largest population within a fifteen weeks evaluation period.
• Figure 5: The evolution of spatial dependencies produced by MSGNN (plotted as colored and filled lines) alongside with weekly national incident number (plotted as dark line).