Deep learning models for forecasting dengue fever based on climate data in Vietnam.

Area of Science:

• Environmental health
• Epidemiology
• Machine learning

Background:

• Dengue fever (DF) poses a significant and escalating health threat in Vietnam, exacerbated by climate change.
• Developing an early-warning system for DF is a key climate change adaptation strategy in Vietnam.

Purpose of the Study:

• To develop an accurate dengue fever prediction model for Vietnam using meteorological factors.
• To inform public health responses for outbreak prevention under future climate change scenarios.

Main Methods:

• Compared deep learning models (CNN, Transformer, LSTM, LSTM-ATT) with traditional machine learning for weather-based DF forecasting.
• Utilized lagged DF incidence and meteorological variables (temperature, humidity, rainfall, evaporation, sunshine) for 20 Vietnamese provinces (1997-2013 training, 2014-2016 testing).
• Evaluated models using Root Mean Square Error (RMSE) and Mean Absolute Error (MAE).

Main Results:

• Attention-enhanced LSTM (LSTM-ATT) demonstrated superior performance in forecasting DF incidence.
• LSTM-ATT outperformed standard LSTM in most provinces for both RMSE and MAE.
• The model accurately predicted DF incidence and outbreak months up to three months in advance.

Conclusions:

• Deep learning models, particularly LSTM-ATT, are effective for meteorological factor-based DF forecasting.
• LSTM-ATT shows potential for DF mitigation strategies and managing other climate-sensitive diseases.