Climate-Driven Advanced Machine Learning Approach for Dengue Incidence Forecasting in Bangladesh

Area of Science:

• Epidemiology
• Environmental Health
• Data Science

Background:

• Dengue fever poses a growing public health challenge in Bangladesh.
• Existing research often lacks long-term scope or comparative modeling, leaving gaps in understanding transmission dynamics and forecasting accuracy.

Purpose of the Study:

• To analyze long-term dengue transmission trends in Bangladesh.
• To compare the predictive validity of statistical and machine-learning forecasting models.
• To identify key climatic predictors of dengue incidence.

Main Methods:

• Analysis of 14 years (2010-2024) of monthly dengue incidence data.
• Inclusion of meteorological variables: temperature, precipitation, and relative humidity.
• Comparative evaluation of four forecasting models: Negative Binomial Regression (NBR), XGBoost, Long Short-Term Memory (LSTM), and Seasonal Autoregressive Integrated Moving Average with Exogenous Variables (SARIMAX).

Main Results:

• Dengue exhibits a predictable seasonal cycle peaking in Q3, overlaid with inter-annual variability.
• Optimal transmission occurs within a climatic envelope: 26°C-30°C, 200-600mm rainfall, and >75% humidity.
• Machine-learning models showed overfitting; SARIMAX demonstrated superior robustness and accuracy (test MAE: 17.0).

Conclusions:

• Model robustness is more critical than algorithmic complexity for dengue forecasting in Bangladesh.
• Long-term forecasts predict a hyperendemic state (2025-2034) with over 2.3 million cases.
• Urgent shift from reactive to proactive surveillance and health system preparedness is needed.