A Cluster-Aggregate-Pool (CAP) ensemble algorithm for improved forecast performance of influenza-like illness.

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• Seasonal influenza leads to significant hospitalizations and deaths in the U.S.
• Influenza-like illness (ILI) forecasts aid public health planning.
• Ensemble forecasts aim to improve accuracy and calibration over individual models.

Purpose of the Study:

• To introduce a novel Cluster-Aggregate-Pool (CAP) ensemble algorithm for influenza forecasting.
• To address issues of forecast similarity and non-identifiability in existing ensemble methods.
• To enhance the accuracy and calibration of influenza-like illness (ILI) predictions.

Main Methods:

• Developed the Cluster-Aggregate-Pool (CAP) ensemble algorithm.
• Clustered individual ILI forecasts, aggregated them into cluster forecasts, and pooled cluster forecasts.
• Evaluated the CAP algorithm on 7 seasons of ILI data from 27 FluSight project models.

Main Results:

• The CAP ensemble improved calibration by approximately 10% compared to non-CAP approaches.
• The CAP ensemble maintained similar accuracy to non-CAP ensemble methods.
• The CAP algorithm offers a generalized framework for ensemble forecasting and provides additional insights into peak influenza timing.

Conclusions:

• The CAP algorithm represents a significant advancement in influenza forecasting ensemble methods.
• This novel approach enhances prediction reliability and offers flexibility for public health officials.
• CAP provides a robust framework for future ensemble modeling in epidemiological forecasting.