Ensemble method for dengue prediction.

Area of Science:

• Epidemiology
• Climate Science
• Mathematical Modeling

Background:

• The 2015 NOAA Dengue Challenge aimed to improve dengue fever forecasting.
• Participants predicted peak dengue case height, peak week, and total cases for Iquitos, Peru, and San Juan, Puerto Rico.
• Standardized input data was provided to all participants for model development.

Purpose of the Study:

• To evaluate the effectiveness of ensemble models in predicting dengue transmission dynamics.
• To compare the performance of ensemble models against other prediction methods in a challenge setting.
• To identify optimal modeling strategies for dengue forecasting.

Main Methods:

• Developed ensemble models by integrating diverse component models: Method of Analogues (dengue and climate data), Holt-Winters (seasonal trends, with/without wavelet smoothing), and historical models.
• Selected best-performing individual component models based on prior data for ensemble construction.
• Created separate ensemble models for each prediction target (peak height, peak week, total cases) and location.

Main Results:

• Ensemble models achieved superior scores for predicting peak dengue height and total case counts in Iquitos, outperforming all other challenge submissions.
• Performance for predicting the peak week of dengue transmission was less successful compared to height and total count predictions.
• The ensemble approach demonstrated significant accuracy for key dengue forecasting indicators in one of the challenge locations.

Conclusions:

• The developed ensemble modeling approach proved highly effective for predicting total dengue cases and peak case height in Iquitos, Peru.
• The study highlights the potential of ensemble methods, combining various data types and models, for accurate dengue forecasting.
• Further refinement is needed for models predicting the timing (peak week) of dengue outbreaks.