Abstract

Dengue fever is a vector-borne disease whose dynamics are substantially driven by climate change and environmental variability. To this end, this study proposed the establishment and assessment of machine learning models for predicting the severity of dengue incidents in Chanthaburi province, Thailand, based on climatic and environmental variables from 2018–2022. Both classification accuracy and confusion matrix analyses were used to compare the implementation of four machine learning algorithms: Random Forest, Gradient Boosting, Extra Trees, and CatBoost. The highest-performing model was able to predict a sum score of 0.50 correctly 86.70% of the time, which suggests that the developed system has good predictive ability, especially in homing in on low-severity dengue cases. Still, challenges associated with recognizing high-severity cases persist. Shapley Additive Explanations (SHAP) sensitivity analysis revealed that specific air pollutant levels (PM10, PM2.5), time-lag parameters, and indicators such as temperature and humidity, particularly during certain periods, were significant predictors of dengue severity. Our results reveal the intricate relationships between environmental variables and the pattern of dengue transmission, arguing for a judicious use of machine learning tools as evidence-based support to inform disease control policies. Future studies that consider other variables, longer time series data, and advanced modeling techniques are needed to increase the accuracy of predictions, especially with respect to improving sensitivity for high-severity dengue outbreaks.