Data-Driven Fine-Grained Epidemic Modeling via Graph Neural Networks Open Access

Abstract

Fine-grained epidemic modeling is crucial for controlling the spread of diseases such as COVID-19. While many graph-based deep learning frameworks for pandemic forecasting achieved powerful performance, seldom use other relevant data sources besides the disease case surveillance data. This paper presents a framework for using the publicly available Social Connectedness Index (SCI) and Social Vulnerability Index (SVI) to enhance the baseline model. These datasets provide valuable insights into the social interactions and socioeconomic status of each location, both potentially significant factors for epidemic spreading dynamics. Experiments were conducted on the U.S. county-level granularity over three datasets with different time frames and geographical scales. We found that SCI and SVI both improve the performance over the original model on some prediction horizons while having comparative performance on other prediction horizons, demonstrating the promising effectiveness of using social- related data sources on pandemic forecasting. Finally, we suggested potential future research directions on data-driven pandemic forecasting. 

Table of Contents

1 Introduction 1

2 Related Work 4

2.1 Graph Neural Networks .................... 4

2.2 Multivariate Time Series Forecasting .................... 5

2.3 Epidemic Modeling .................... 6

3 Problem Definition 7

4 Comparative Study 8

4.1 Background .................... 8

4.1.1 StemGNN .................... 8

4.1.2 Cola-GNN .................... 9

4.1.3 STAN .................... 9

4.1.4 CausalGNN .................... 10

4.2 Qualitative Comparison .................... 10

4.3 Quantitative Comparison .................... 12

4.3.1 Dataset .................... 13

4.3.2 Setup .................... 13

4.3.3 Evaluation Metric .................... 14

4.3.4 Results .................... 15

4.4 Analysis .................... 16

5 Enhancement Study 17

5.1 Dataset .................... 17

5.1.1 Geographical Adjacency .................... 17

5.1.2 Social Connectedness Index .................... 17

5.1.3 Social Vulnerability Index .................... 18

5.2 Cola-GNN: A Closer Look .................... 19

5.2.1 Directed Spatial Influence Learning .................... 20

5.2.2 Multi-Scale Dilated Convolution .................... 21

5.2.3 Graph Message Passing -- Propagation .................... 22

5.2.4 Output Layer -- Prediction .................... 22

5.2.5 Optimization .................... 23

5.3 Proposed Method .................... 23

5.4 Experiment .................... 25

5.5 Analysis .................... 26

6 Conclusion 30

Appendix A Dataset 32

About this Honors Thesis

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Emory College
Department	Computer Science
Degree	B.S.
Submission	Honors Thesis
Language	English
Research Field	Computer Science
Keyword	Epidemic Forecasting Machine Learning Graph Neural Networks
Committee Chair / Thesis Advisor	Xiong, Li, Emory University
Committee Members	El-Sayed, Nosayba, Emory University Züfle, Andreas , Emory University

Last modified

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Data-Driven Fine-Grained Epidemic Modeling via Graph Neural Networks ()	2023-04-17 14:31:10 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions