Improving Dengue Case Prediction in Bandung City using Random Forest and SHAP on Climate Demographic Data

Abstract

Dengue Hemorrhagic Fever remains a major public health concern in urban areas of Indonesia, particularly in Bandung City, due to its fluctuating incidence and strong dependence on environmental and population factors. This study focuses on improving dengue case prediction by integrating climate and demographic data through systematic feature engineering and explainable machine learning based on the Random Forest algorithm. Historical dengue case data from Bandung City were used to develop and evaluate the proposed prediction model. The evaluation results show that the Random Forest model achieved an R² value of 0.9032 and an RMSE of 2.3748, indicating reliable predictive performance and good generalization capability. The applied feature engineering strategy effectively enhanced data representation by capturing temporal dynamics, case growth patterns, and interactions among climate variables. Furthermore, model interpretability was improved through the application of Explainable Artificial Intelligence using SHAP, which revealed that temporal features derived from previous dengue case trends were the most influential factors, followed by climate interaction variables. These findings demonstrate that the proposed approach improves prediction accuracy while providing transparent and epidemiologically meaningful insights to support data driven dengue early warning systems at the regional level.