A disease-agnostic approach to ensemble learning for infectious disease forecasting

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• Accurate infectious disease forecasting is vital for public health interventions, but real-time prediction remains challenging.
• Ensemble modeling, averaging multiple models, improves forecast accuracy but traditionally requires extensive historical data.
• This reliance on historical data makes traditional methods impractical for emerging diseases with limited data.

Purpose of the Study:

• To introduce epiFFORMA, a novel, disease-agnostic ensembling strategy for infectious disease forecasting.
• To develop a method that determines ensemble weights without relying on historical disease-specific data.
• To evaluate epiFFORMA's performance against existing methods and individual component models.

Main Methods:

• Developed epiFFORMA, an ensembling strategy building on the FFORMA model.
• Utilized synthetic data to capture epidemiological dynamics for weight determination.
• Tested epiFFORMA across diverse infectious diseases including COVID-19, dengue, measles, and influenza-like illness.

Main Results:

• epiFFORMA outperformed a naive equal-weighting ensemble strategy in forecasting various infectious diseases.
• The proposed method demonstrated superior performance compared to individual component models on average.
• The disease-agnostic nature of epiFFORMA allows for effective forecasting even with limited or no historical data.

Conclusions:

• epiFFORMA offers a robust solution for infectious disease forecasting, particularly for emerging threats.
• The strategy's ability to function without historical data makes it highly adaptable and practical for public health.
• This approach enhances the reliability of real-time infectious disease outbreak predictions.