Data-Driven Machine Learning-Based Forecasting of Dengue in Bangladesh: Supporting Digital Health Approaches for...

Dengue is a significant vector-borne disease that has severely impacted public health in Bangladesh, underscoring the growing importance of digital health in enhancing surveillance and prevention. Understanding its trends and future estimates is crucial for improving early prevention strategies. This study aimed to model trends and select the best model to forecast dengue cases in Bangladesh for the next 5 years to aid digital health early warnings.

The monthly dengue case data (January 2000 to December 2023) were obtained from the Directorate General of Health Services (DGHS). An autoregressive integrated moving average (ARIMA) and eXtreme gradient boosting (XGBoost) model were employed to analyze the data. The root mean square error (RMSE), mean absolute error (MAE), and mean absolute scaled error (MASE) were used to evaluate the model's performance.

From 2000 to 2023, Bangladesh reported 565,890 confirmed dengue cases, marking a sharp peak in 2023 with 321,179 cases, alongside high incidence rates of 194.47, and a lowest count in 2014 with only 375 cases. A distinct seasonal trend was observed, with cases rising in June, peaking in August, and declining by October. To identify the most suitable model, both ARIMA and XGBoost were evaluated. Performance metrics indicated that XGBoost outperformed ARIMA (RMSE = 0.63, MAE = 0.54, MASE = 0.39) in predicting dengue cases. Feature importance analysis showed that recent dengue incidence, especially at lag 1, was the most prominent predictor, with further impact from longer-term and seasonal recurrence patterns. Consequently, XGBoost was employed to forecast future incidences, projecting that dengue cases may range from 35,297 to 330,242 between 2024 and 2028, suggesting a potential rise in future outbreaks.

The findings underscore the urgent need to strengthen early warning systems and leverage digital health tools to manage the escalating dengue threat. Integrating machine learning models into public health strategies can enhance predictive accuracy and inform targeted interventions. Future research should consider additional factors, such as climate and urbanization, to refine projections and support effective disease management.