Spatial-Temporal Synchronous Graph Transformer network (STSGT) for COVID-19 forecasting.

Area of Science:

• Epidemiology
• Data Science
• Network Science

Background:

• COVID-19 poses a significant global health and economic threat.
• Accurate forecasting of pandemic dynamics is crucial for effective public health response.
• Existing spatial-temporal forecasting models struggle to capture complex dependencies in time-series data.

Purpose of the Study:

• To introduce a novel Spatial-Temporal Synchronous Graph Transformer network (STSGT) for COVID-19 time-series forecasting.
• To effectively capture complex spatial and temporal dependencies in pandemic data.
• To improve the accuracy of forecasting future pandemic status.

Main Methods:

• Developed STSGT, integrating Graph Convolutional Networks (GCN) with transformer self-attention mechanisms.
• Utilized a synchronous spatial-temporal graph to model dynamic patterns and interdependencies.
• Employed GCN and transformer layers to capture evolving patterns in COVID-19 time-series data.

Main Results:

• STSGT significantly outperformed state-of-the-art algorithms in spatial-temporal forecasting tasks.
• Achieved an average improvement of 12.19% in Mean Absolute Error (MAE) for infected cases and 3.42% for death cases over a 12-day horizon.
• Demonstrated superior performance in forecasting daily infected cases at both state and county levels in the US.

Conclusions:

• STSGT offers a powerful new approach for accurate pandemic forecasting.
• The model's ability to capture complex spatial-temporal dependencies enhances prediction accuracy.
• STSGT provides a valuable tool for public health officials in managing evolving pandemics.