Data-Driven Dengue Prevention Strategies in Bangladesh using Explainable Artificial Intelligence and Causal Inference

Abstract

Dengue is one of the major public health problems in Bangladesh. This country experiences periodic dengue outbreaks and it is particularly vulnerable to climate-related factors. The increasing dengue burden in the country necessitates reliable prediction-based instruments to inform public health strategies. In this study, we employed high-performance machine learning (XGBoost), combined with explainable artificial intelligence (XAI). This evaluates the predictive relevance of climate, socioeconomic, healthcare, and land-use variables for dengue incidence in Bangladesh from January 2010 to December 2024. We also applied causal inference (CAI) techniques by using the DoWhy framework. It enables robust interpretation of the variables’ impact and their directional influence on dengue incidence. The model performance was assessed using RMSE, MAE, and MAPE matrices. Dengue cases increased suddenly in Bangladesh during 2019, 2023, and 2024, with each year exceeding 100,000 total cases and monthly peaks of 50,000 cases. The XGBoost model was the top-performing model (RMSE: 11,365.2; MAE: 7,014.68; MAPE: 764.36). It offers the highest predictive accuracy of dengue risk. Climate indicators were the strongest contributor to dengue prediction (41.87%), followed by sociodemographic (35.03%), healthcare (19.16%), and landscape factors (3.95%). Our findings show that machine learning, specifically XGBoost, can effectively predict dengue cases and help identify key ecological and structural drivers. It highlights key ecological and structural drivers, supporting the integration of real-time, multisectoral data into public health planning and decision-making. It also improved digital literacy, health infrastructure, and early warning systems, which are essential for mitigating dengue outbreaks in Bangladesh and similar endemic countries.

International Journal of Statistical Sciences, Vol. 25(2), November, 2025, pp 101-113