Leveraging Graph-Based Deep Learning for Epidemic Modeling

Accurate prediction and prompt identification of contagious diseases are very crucial for protecting the global economy and public health. This study leverages the power of a Graph Autoencoder (GAE) to model disease epidemics while blending socioeconomic and demographic information along with statistical data. Traditional techniques have the drawback that they cannot capture the complex nonlinear and dynamic relationship of various factors, in a streaming scenario. In this work, we create month-by-month attributed graphs where each node represents a country or region, incorporating socioeconomic and demographic data alongside COVID-19 pandemic statistics. Each edge represents the relationships between these countries or regions. We then perform graph clustering to analyze patterns and connections within this data. Subsequently, we employ emerging deep learning methods like Graph autoencoder (GAE) with GCN, GAE with GAT, and EvolveGCN-H on link prediction tasks to identify relationships among similar countries. Experimental results demonstrate that EvolveGCN-H outperforms others in offering better prediction and thus providing a robust tool for global surveillance and monitoring.