Pandemic velocity: Forecasting COVID-19 in the US with a machine learning & Bayesian time series compartmental model

Area of Science:

• Epidemiology
• Biostatistics
• Computational Biology

Background:

• Accurate COVID-19 pandemic forecasting is crucial for public health decision-making.
• Predicting disease spread is challenging due to novel virus characteristics and evolving societal factors.

Purpose of the Study:

• To develop and evaluate a sophisticated epidemiological model for predicting COVID-19 case growth and mortality in U.S. states.
• To provide reliable daily projections and uncertainty estimates for pandemic trajectories.

Main Methods:

• Integrated a Bayesian time series model and a random forest algorithm within an epidemiological compartmental model.
• Modeled case velocity using location-specific curves and Bayesian inference, incorporating prior information.
• Predicted deaths using a random forest trained on COVID-19 data and population characteristics.

Main Results:

• The model demonstrated predictive accuracy over 21-day forecasts when trained on increasing periods of pandemic data.
• Significant variations in predicted trajectories and uncertainty were observed across different U.S. states.
• The model successfully generated daily projections and interval estimates for COVID-19 cases and deaths.

Conclusions:

• The developed model offers sophisticated and accurate COVID-19 predictions with uncertainty quantification for U.S. states.
• This framework provides a robust platform for ongoing pandemic forecasting, adaptable to changing public health responses.