Enhancing Epidemic Forecasting: Evaluating the Role of Mobility Data and Graph Convolutional Networks
Suhan Guo, Zhenghao Xu, Furao Shen, Jian Zhao · 2025
Read Enhancing Epidemic Forecasting: Evaluating the Role of Mobility Data and Graph Convolutional Networks by Suhan Guo, Zhenghao Xu, Furao Shen, Jian Zhao (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-11028.txt.
Introduction
Forecasting epidemics would seem to be a natural job for mobility data, since disease rides on movement. Suhan Guo and co-authors test that assumption in two phases. A pilot study first checked whether mobility data matter at all. Then they added graph convolutional networks onto a transformer backbone and measured the change. Such networks model spread across a map of connected places.
The result surprises: neither mobility data nor the graph modules clearly improved the forecasts. Adding mortality and hospitalisation data, by contrast, markedly raised accuracy. Maps derived from the graph networks did line up notably with lockdown orders, suggesting such maps can act as sensitive indicators of mobility even when they do not sharpen predictions. The authors offer the study as a corrective for decision-makers and modellers who assume that movement data automatically help.
Catalog record
| CREATOR | Suhan Guo, Zhenghao Xu, Furao Shen, Jian Zhao |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 60 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-11028
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Enhancing Epidemic Forecasting: Evaluating the Role of Mobility Data and Graph Convolutional Networks Suhan Guo, Zhenghao Xu, Furao Shen, Jian Zhao Abstract Accurate prediction of contagious disease outbreaks is vital for informed decision-making. Our study addresses the gap between machine learning algorithms and their epidemiological applications, noting that methods optimal for benchmark datasets often underperform with real-world data due to difficulties in incorporating mobility information. We adopt a two-phase approach: first, assessing the significance of mobility data through a pilot study, then evaluating the impact of Graph Convolutional Networks (GCNs) on a transformer backbone. Our findings reveal that while mobility data and GCN modules do not significantly enhance forecasting performance, the inclusion of mortality and hospitalization data markedly improves model accuracy. Additionally, a comparative analysis between GCN-derived spatial maps and lockdown orders suggests a notable correlation, highlighting the potential of spatial maps as sensitive indicators for mobility. Our research offers a novel perspective on mobility representation in predictive modeling for contagious diseases, empowering decision-makers to better prepare for future outbreaks. keywords COVID-19 incidence forecasting, mobility representation, Graph Convolutional Networks (GCN), spatial-temporal neural network Introduction Accurate forecasting of daily case numbers for infectious diseases is crucial for disease management. The exponential growth of cases serves as an indicator of upcoming waves and the emergence of new variants [ 1 ] . Such forecasts allow governments to update and implement directives, facilitating proactive resource allocation to prevent strain on healthcare systems. Airborne infectious disease outbreaks heavily rely on frequent human interaction that large-scale social activities significantly amplifying the transmission of communicable diseases [ 2 , 3 , 4 , 5 , 6 ] . The global impact of SARS-CoV-2 is unparalleled compared to previous coronavirus pandemics [ 7 ] , owing to its highly contagious nature and the increased accessibility of both intra- and inter-national travel facilitated by enhanced global connectivity [ 8 ] . International travels across continents promote the dissemination of pathogens, ultimately resulting in pandemics [ 9 , 10 ] . Extensive research has demonstrated that human mobility plays a pivotal role in driving the geographic spread of infectious diseases. Consequently, the routine movement of individuals between sub-regions is intertwined with their epidemic phases, necessitating a spatial perspective in disease forecasting [ 11 , 12 , 13 , 14 ] . The use of mobility data as a proxy for effective contact has raised concerns regarding its efficacy in capturing its correlation with increased disease transmission [ 15 ] . Nevertheless, mobility data remains a crucial source for modeling disease transmission in government surveillance [ 16 , 17 ] and epidemic research [ 18 , 19 , 20 , 21 ] . Real-time behavioral data, such as surveys on social media platforms and search term volumes, are employed to describe mobility patterns, yet their collection poses challenges due to the inherent difficulty in tracking human behavior [ 22 ] . The resulting datasets often exhibit limited geographical coverage and contain numerous missing data points, thereby diminishing the effectiveness of mobility data as a proxy for human movements. Accurately forecasting the spatial-temporal dynamics of infectious diseases requires considering mobility patterns and capturing local trends. Many models rely solely on historical daily case data for predictions [ 23 , 24 , 25 ] . However, this univariate analysis overlooks the spatial information and clinical outcomes of neighboring regions or countries, which can greatly aid in estimating local future cases [ 26 ] . Previous works utilizing the neural networks, featuring a temporal backbone and a GCN module, present a promising alternative by explicitly capturing associations in multivariate time series data [ 27 , 28 , 29 , 30 ] . This innovative approach effectively models the spatial-temporal structure of SARS-CoV-2 incidence, leading to more accurate forecasting than conventional epidemiological tools. To enhance the representation of human movements with neural networks, we suggest assessing the influence of mobility data on forecasting tasks. In our pilot study, we compare the performance of models that only use incidence data with those that incorporate both incidence and mobility information. Recognizing the inherent challenges of missing data and geographical coverage limitations in mobility data, we advocate for alternative data sources to better capture mobility in forecasting tasks. The relationship among clinical indicators—incidence, hospitalization rate, and mortality—is mutually dependent, as depicted in Supplementary Fig.2. Inspired by Multi-task Learning (MTL) [ 31 ] , which improves generalization by sharing representations among related tasks to mitigate data sparsity and overfitting risks [ 32 ] , we suggest that incorporating mortality and hospitalization rates with incidence data can help the model infer mobility information from clinical data rather than relying solely on mobility data. Beyond data, adjusting model structures offers another pathway for improving forecasting results. In our main experiments, we aim to evaluate the contribution of the GCN module in capturing human movement without explicitly relying on mobility data. Previous studies have explored various temporal backbones, such as LSTM [ 18 , 30 ] , GRU [ 33 ] , CNN [ 29 ] , SIR models [ 28 ] , and GCN modules, to represent spatial information during the pandemic. However, due to differences introduced by temporal structures, the effectiveness of the GCN module remains inadequately assessed. Therefore, we opt for the powerful transformer as the temporal backbone to evaluate the effectiveness of incorporating a GCN module with physical-distance-based and dynamically derived adjacency matrices. To study the influence of cross-border commuting, we focus on two distinct regions: the European Union (EU), representing countries with interconnected mobility, and the United States (US), reflecting domestic mobility patterns. For our analysis, we have chosen 28 EU countries and 49 US states based on data availability. These regions serve as external validation for each other in developing a forecasting model for SARS-CoV-2. Our study aims to identify the primary contributor to enhancing forecasting capabilities by aiding in mobility retrieval. We propose using a transformer backbone to model temporal correlation, while spatial correlation is captured by a GCN module. The adjacency matrices include a physical-distance-based matrix generated using coordinates and a non-distance-related one generated by an attention mechanism with a normal truncating constraint. Additionally, by incorporating supplementary data across sub-regions, we test the contribution of MTL in improving predictive performance. A sensitivity analysis is carried out, comparing the prediction performance of the transformer backbone with that of the ARIMA model for univariate forecasting and tree-based and linear models for multivariate forecasting. Figure 1 : Research Pipeline The project consists of four major steps, termed “preprocess”, “train”, “evaluate” and “infer”. The first three steps serve as pipelines to answer the questions posed in the “infer” step. Results Research Pipeline Our study aims to find the core contributor to the spatiotemporal neural networks for integrating mobility information into SARS-CoV-2 incidence forecasting. This objective is divided into three questions, outlined in the “Infer” section of Fig. 1 . The first question examines the necessity of mobility data in the incidence forecasting task, which will be addressed in the pilot study. This initial investigation will provide crucial insights to support subsequent experiments. Both the pilot study and the main experiment will delve into the possibility of extracting mobility information from related data. Building upon the findings of