Improving dengue fever predictions in Taiwan based on feature selection and random forests

Area of Science:

• Environmental Science
• Public Health
• Epidemiology

Background:

• Dengue fever is a significant vector-borne disease in tropical and subtropical regions.
• Existing dengue prediction models in Taiwan lack air quality index (AQI) integration.
• Investigating AQI's role in dengue occurrence is crucial for public health.

Purpose of the Study:

• To develop a dengue fever prediction model integrating meteorological factors, vector index, and AQIs.
• To evaluate the impact of novel variables like PM2.5 and UV index on dengue prediction.
• To compare the performance of various machine learning algorithms for dengue forecasting.

Main Methods:

• Collected 805 meteorological records from 2013-2015.
• Incorporated AQIs, particulate matter (PM10, PM2.5), and UV index into machine learning models.
• Utilized random forests, achieving an AUC of 0.9547 on the test set.

Main Results:

• Random forests outperformed other algorithms in dengue prediction.
• Temperature was the most significant factor, with a notable effect below 30°C.
• AQIs showed a negative effect on dengue fever occurrence, despite lower importance than temperature.

Conclusions:

• This study is the first to show a negative impact of AQI on dengue fever in Taiwan.
• The developed prediction model can serve as an early warning system for dengue outbreaks.
• Integrating AQI into prediction models enhances public health strategies against dengue.