Real-time forecasting of epidemic trajectories using computational dynamic ensembles

Area of Science:

• Epidemiology
• Computational Biology
• Mathematical Modeling

Background:

• Forecasting infectious disease spread is complex due to data and model uncertainties.
• Existing models often struggle to accurately predict epidemic trajectories.
• There is a need for robust forecasting methods that quantify uncertainty.

Purpose of the Study:

• To develop and evaluate an ensemble model for sequential forecasting of social dynamic processes, specifically infectious disease outbreaks.
• To assess the uncertainty of the ensemble model using a frequentist computational bootstrap approach.
• To compare the performance of the ensemble model against individual models and established forecasting challenges.

Main Methods:

• Developed an ensemble model by weighting a set of plausible phenomenological models, including the Generalized-Growth Model (GGM) and Generalized Logistic Model (GLM).
• Employed a frequentist computational bootstrap approach to evaluate forecast uncertainty.
• Utilized the root-mean-square error (RMSE) for model calibration and forecasting performance evaluation.
• Applied the model to outbreak scenarios from the Ebola Forecasting Challenge.

Main Results:

• The ensemble model, combining GGM and GLM, demonstrated superior overall mean RMSE performance compared to individual models and participants in the Ebola Forecasting Challenge.
• The ensemble model consistently provided more accurate forecasts than the GGM and GLM individually across various forecasting horizons.
• Forecast accuracy varied by horizon; the ensemble model excelled at 2-3 week horizons, while GLM performed better at 1 and 4 weeks.

Conclusions:

• Ensemble modeling offers a promising approach to enhance the accuracy and reliability of infectious disease spread forecasts.
• Accounting for model uncertainty through methods like bootstrapping is crucial for robust epidemic forecasting.
• The developed GGM-GLM ensemble model shows potential for improving short-term epidemic outbreak predictions.