Climate-driven dengue forecasting in Bangladesh: division-specific feature-set design and lag structure

Abstract Bangladesh shows marked annual variation in dengue incidence, partly driven by meteorological effects on Aedes breeding and transmission. We contrast consistently high-burden Dhaka with Barishal, where dengue incidence has recently increased, and emphasize feature-set design and predictor structure as the main methodological contributions. Using monthly dengue data from Directorate General of Health Services (DGHS, 2025 Daily dengue status report) and meteorological data from WorldWeatherOnline (World Weather Online. https://www.worldweatheronline.com/bangladesh-weather.aspx) for January 2022–October 2025, we compare four climate feature sets that vary wetness (rainy days versus rainfall) and sunshine (sun days versus sun hours), with temperature and humidity included in all sets. Correlation analysis across 0–4-month climate lags shows that rainfall has the strongest positive correlations at a 2-month lag, humidity at a 1-month lag, sunshine has the strongest negative correlations at a 2-month lag and temperature shows weak positive correlations at longer lags. For forecasting, we evaluate lagged climate covariates alone and combined with 1-month lagged dengue incidence. We compare multivariate Poisson regression (MPR), artificial neural networks (ANNs), extreme gradient boosting (XGBoost) and Seasonal Autoregressive Integrated Moving Average with eXogenous (SARIMAX) regressors. For Dhaka, ANN-1 with SET-1 performs best (RMSE = 2176.70), whereas for Barishal, SARIMAX((0,1,1)(1,0,0,12)) with SET-2 performs best (RMSE = 817.56). Analyses use consistent monthly aggregation and division-specific tuning.