District-Level Dengue Early Warning Prediction System in Bangladesh Using Hybrid Explainable AI and Bayesian Deep...

Area of Science:

• Epidemiology
• Public Health
• Data Science

Background:

• Dengue poses a significant public health challenge in Bangladesh, with increasing case numbers.
• Existing surveillance models struggle to address district-level disparities in dengue transmission.
• Climatic variability, urbanization, and socio-economic factors influence dengue spread.

Purpose of the Study:

• To develop a district-level dengue early warning system for Bangladesh.
• To integrate diverse determinants including climate, socio-demographic, economic, healthcare, and environmental data.
• To generate accurate, interpretable, and actionable dengue outbreak predictions.

Main Methods:

• Utilized machine learning (MLP) and deep learning (ConvLSTM) models.
• Incorporated explainable AI (SHAP) for model interpretability.
• Employed Bayesian spatio-temporal models (BYM2_RW2) to account for spatial and temporal dependencies.

Main Results:

• Climate factors (humidity, minimum temperature, rainfall) were key predictors of dengue transmission.
• Socio-economic factors like poverty and healthcare capacity (nursing density) also significantly contributed to predictions.
• The MLP model excelled in yearly predictions (accuracy=0.93), while ConvLSTM performed best for monthly forecasts (recall=0.88).

Conclusions:

• The developed integrated framework provides transparent and interpretable district-level dengue early warnings.
• This system supports adaptive strategies for dengue outbreak preparedness and optimizes resource allocation in Bangladesh.
• The study highlights the importance of multi-factorial data integration for effective dengue surveillance.