Early Detection of Dengue Hemorrhagic Fever Using Patient Medical Data with Ensemble Learning Methods

Abstract

Dengue Hemorrhagic Fever (DHF) remains a major public health concern in Indonesia and worldwide, where delayed diagnosis increases the risk of severe complications and mortality. Conventional laboratory-based diagnostics are time-consuming and often less accessible in resource-limited healthcare settings. This study aims to develop an early detection model for DHF using only initial clinical symptoms and demographic data extracted from electronic medical records at RSUD Brigjend H. Hasan Basry Kandangan. A total of 649 patient records (352 DHF cases and 297 non-dengue) were analyzed using the CRISP-DM framework. Five ensemble learning algorithms Random Forest, Bagging, AdaBoost, and Gradient Boosted Tree were evaluated across 80:20, 70:30, and 60:40 data splits and validated using 5-fold and 10-fold cross-validation. Random Forest consistently delivered the best and most stable performance, achieving up to 90.00 % accuracy and 0.967 AUC in the 80:20 split and mean accuracies of 88.91 % (5-fold) and 88.29 % (10-fold) in cross-validation. Further hyperparameter tuning enhanced model stability and prevented overfitting. The findings confirm that initial clinical symptoms and demographic attributes can reliably identify DHF cases early, enabling faster and more affordable screening prior to laboratory confirmation. This machine learning based decision-support model has the potential to significantly improve early clinical management of dengue fever.