Ensemble approaches for short-term dengue fever forecasts: A global evaluation study.

Area of Science:

• Tropical medicine
• Epidemiology
• Computational modeling

Background:

• Dengue fever is a significant global health threat, causing widespread hospitalization and mortality, particularly in tropical regions.
• Effective dengue surveillance and forecasting are crucial for implementing timely public health interventions and resource allocation.
• Existing forecasting models show variable performance, highlighting the need for improved predictive accuracy.

Purpose of the Study:

• To develop and evaluate ensemble modeling approaches for forecasting dengue fever cases 1 to 3 months ahead.
• To assess the predictive performance of these ensemble models across diverse geographical locations and timeframes.
• To compare the efficacy of ensemble models against individual mechanistic, statistical, and machine learning models.

Main Methods:

• Ensemble modeling integrating mechanistic, statistical, and machine learning approaches.
• Retrospective and prospective out-of-sample validation across over 180 global locations.
• Real-time forecasting platform evaluation during 2022-2023, accounting for data limitations.

Main Results:

• Ensemble models demonstrated improved predictive performance compared to individual models across various locations and time periods.
• No single component model consistently outperformed others, underscoring the value of ensemble approaches.
• Ensemble models consistently ranked among the top three performers in both retrospective and prospective evaluations.

Conclusions:

• Ensemble modeling offers a robust strategy for enhancing dengue fever forecasting accuracy and reliability.
• These improved forecasts can aid decision-makers in anticipating healthcare demands and strengthening public health preparedness amidst uncertainty.
• The proposed ensemble approach represents a significant advancement over previous dengue forecasting efforts.