Developing a dengue forecast model using machine learning: A case study in China.

Area of Science:

• Epidemiology
• Public Health
• Machine Learning

Background:

• Dengue fever poses a significant public health challenge in China, with recent increases in incidence and geographic spread.
• Existing dengue incidence forecasts in China lack accuracy and timeliness.
• There is a need for advanced predictive models to address dengue surveillance.

Purpose of the Study:

• To develop an accurate dengue incidence predictive model using state-of-the-art machine learning algorithms.
• To evaluate the performance of various machine learning models for dengue forecasting in China.

Main Methods:

• Collected weekly dengue cases, Baidu search queries, and climate data (temperature, humidity, rainfall) from 2011-2014 in Guangdong.
• Constructed a dengue search index and integrated it with climate factors and temporal variables for model development.
• Compared multiple machine learning algorithms, including Support Vector Regression (SVR), Gradient Boosted Regression Trees (GBM), and others, assessing performance using RMSE and R-squared.
• Validated models using data from five additional Chinese provinces.

Main Results:

• The Support Vector Regression (SVR) model, selected via cross-validation, accurately forecasted dengue epidemics, including the peak of a major 2014 outbreak.
• The SVR model demonstrated consistently lower prediction errors compared to other models for tracking dengue dynamics and forecasting outbreaks.
• Model residuals were analyzed using autocorrelation and partial autocorrelation functions to ensure validity.

Conclusions:

• The Support Vector Regression (SVR) model significantly outperformed other evaluated forecasting techniques for dengue incidence.
• These findings provide a valuable tool for governmental and community-level early response to dengue epidemics.
• The developed model enhances the ability to predict and manage dengue outbreaks in China.