Predicting Seasonal Influenza Hospitalizations Using an Ensemble Super Learner: A Simulation Study

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• Influenza forecasting is crucial for outbreak response.
• Existing efforts primarily focus on influenza-like activity, not hospitalizations.
• Accurate prediction of influenza hospitalizations remains a challenge.

Purpose of the Study:

• To evaluate a super learner's predictive performance for influenza hospitalizations.
• To assess predictions of peak hospitalization rate, peak week, and cumulative rate.
• To compare ensemble machine learning against individual algorithms and naive predictions.

Main Methods:

• A simulation study using 15,000 simulated influenza hospitalization curves.
• Training an ensemble machine learning algorithm for weekly predictions.
• Comparing ensemble, best individual algorithm, and naive (median) predictions.

Main Results:

• Ensemble predictions matched naive predictions early in the season.
• Predictive accuracy consistently improved as the season progressed.
• Individual algorithm performance varied weekly, often similar to the ensemble.

Conclusions:

• An ensemble super learner significantly improved influenza hospitalization predictions compared to naive methods.
• The ensemble approach showed enhanced forecasting as the season advanced.
• Future work should incorporate empirical data and prospective probabilistic forecasting.