Improving the Accuracy of Early Diagnosis of Dengue Hemorrhagic Fever Based on Clinical Symptoms Using Random Forest

Abstract

The development of machine learning in the field of health provides important opportunities for improving the accuracy of disease diagnosis, including dengue hemorrhagic fever (DHF), which remains a major health problem in Indonesia. This study aims to develop an early DHF diagnosis model based on the Random Forest algorithm using clinical symptom data from patients at the Kosasih Group Clinic. The research was conducted using a quantitative approach through the CRISP-DM stages, which included data acquisition, validation, cleaning, and preprocessing, covering missing value handling, normalization, and class imbalance management using SMOTE. The dataset was then divided using stratified sampling to maintain class proportions, followed by training the Random Forest model optimized using Bayesian Optimization to obtain the best combination of hyperparameters. Performance evaluation was carried out using accuracy, precision, recall, F1-score, and ROC-AUC metrics, and validated using stratified k-fold cross-validation to ensure model stability. Model interpretability was analyzed using SHAP and LIME to identify the contribution of each clinical symptom to the prediction. The results showed that the model was able to provide high classification performance, with increased sensitivity to DHF cases after applying SMOTE and consistent interpretation of clinical symptoms such as fever, joint pain, and nausea. These findings confirm the potential of Random Forest as a reliable model to support the development of AI-based clinical decision support systems (CDSS) for early diagnosis of DHF in primary health care facilities.