Combinations of Optimization Method and Balancing Technique in Hypertension Classification with Machine Learning

Hypertension is a condition in which blood vessels experience continuous pressure higher than normal limits which can cause pain and even death. Hypertension is classified into several classes based on the measured blood pressure. To correctly diagnose hypertension is a critical task that requires medical specialists who are unfortunately not evenly distributed in every region. This research aims to implement Particle Swarm Optimization for hyperparameter tuning in machine learning algorithms in hypertension disease classification. The approach was developed by comparing the performance of Random Forest (RF), Light Gradient Boosting Machine (LGBM), and Extra Trees (ET). Each algorithm was trained using its default hyperparameters, tuned with Grid Search and Cross-validation (GSCV), and the Particle Swarm Optimization with Cross-validation (PSO-CV). We consider recall to be the primary evaluation metric due to the imbalance in the dataset. The experiment results show that the combination of the LGBM and PSO-CV is the best combination of algorithm and hyperparameter optimization method with precision, recall, F1-score, ROC-AUC, and PR-AUC values of 0.22, 0.63, 0.33, 0.79, and 0.24, respectively. The results of this study prove that PSO might positively influence model performance, particularly in the case of unbalanced data.