Addressing Imbalance Data Label Distribution in Dengue Severity Classification with Hybrid Machine Learning Approach

Early detection of disease is essential for ensuring proper diagnosis and treatment, which helps to reduce complications and limit the spread of cases. Dengue fever remains a major public health issue in Indonesia, with some regions reporting case fatality rates above 1%. This study aims to develop a robust classification model to predict the severity of dengue fever based on laboratory data, addressing the issue of imbalanced label distribution. To achieve this, we applied a combination of oversampling, machine learning, and optimization techniques. Specifically, we utilized the SMOTE-ENN method to enhance the representation of minority classes and applied XGBoost for multi-class classification. Additionally, Particle Swarm Optimization (PSO) was used to fine-tune the model’s hyperparameters. After testing 15 experimental scenarios, the hybrid SMOTE-ENN-XGBoost-PSO approach delivered the best performance, achieving an accuracy of 95% and an F1-score of 94%. These results demonstrate that the proposed approach effectively handles data imbalance and improves predictive accuracy. This research serves as a foundation for future work in applying advanced algorithms and Federated Learning to improve multi-class classification in healthcare.