Introduction to Forecasting Infectious Disease Outbreaks

Meeting Times:

• Wednesday, July 15, 9:00 AM – 5:00 PM
• Thursday July 16, 9:00 AM – 5:00 PM
• Friday July 17, 9:00 AM – 12:00 PM

Classroom: TBA

Module Summary:
This module introduces the fundamentals of infectious disease forecasting, from building simple and more complex predictive models to evaluating and combining them in real-world collaborative settings. Participants will learn to construct probabilistic forecasts, critically assess forecast quality using proper scoring rules, and leverage ensemble methods to improve predictive performance. The module also covers the practical infrastructure of forecast hubs, which are standardized platforms for collecting, combining, and evaluating forecasts from multiple models, either for a collaborative modeling project with multiple modeling contributors or for a smaller experimental model testbed that you might run yourself on your laptop. The module also demonstrates how forecasting can be integrated with nowcasting approaches to handle incomplete data in real-time outbreak analysis.

Prerequisites:
It is expected that course participants have basic knowledge of statistics and mathematics and rudimentary knowledge of infectious disease epidemiology. It is also expected that participants have basic computer software knowledge and have at least a basic familiarity with the software R.

Module Content:

• Forecasting fundamentals: Introduction to epidemiological forecasting using regression-based models, including data visualization, transformations, and handling seasonality in time-series data
• Forecast evaluation: Assessing forecast quality through calibration, accuracy, bias, and sharpness using proper scoring rules and visualization techniques
• Ensemble forecasting: Strategies for combining multiple models to improve predictive performance, including quantile-based forecast representations
• Forecast hubs: Working with collaborative forecasting infrastructure using hubverse tools for standardized forecast submission, evaluation, and ensemble creation
• Integrating nowcasting and forecasting: Approaches for making predictions when recent data are incomplete due to reporting delays, connecting real-time estimation with forward projection