A Stacking Ensemble Prediction Model for the Occurrences of Major Adverse Cardiovascular Events in Patients With Acute Coronary Syndrome on Imbalanced Data

The major adverse cardiovascular events (MACE) often occur with high morbidity and mortality globally. It is very important to predict the MACE occurrences accurately in patients with acute coronary syndrome (ACS). Therefore, this paper proposes a stacking ensemble model for the prediction of MACE occurrences in patients with ACS at early stage. Our research contents are summarized as follows. First, we use the Korea Acute Myocardial Infarction Registry National Institutes of Health (KAMIR-NIH) dataset, and our experimental data are extracted from the raw data and preprocessed. Second, we apply three data sampling approaches, such as borderline synthetic minority oversampling technique (Borderline-SMOTE1), cluster centroids undersampling, and synthetic minority oversampling techniques (SMOTE) plus Tomek Links (SMOTETomek) hybrid technique, to solve the class imbalance problem. Third, to develop a stacking ensemble prediction model for the occurrences of MACE, we apply seven widely used machine learning algorithms, such as logistic regression (LR), support vector machine (SVM), K-Nearest Neighbors (KNN), decision tree (DT), random forest (RF), extreme gradient boosting (XGBoost), and adaptive boosting (AdaBoost), as base learners. Fourth, the performance of the proposed stacking ensemble model is compared with the seven base learners using the three data sampling techniques. In the result, the proposed stacking ensemble model with the SMOTETomek shows the best performance with accuracy of 0.9862, precision 0.9976, recall 0.975, F1-score 0.9862, g-mean 0.9863, and AUC 0.9863 and provided a better solution for imbalanced dataset. Consequently, our finding is that the proposed stacking ensemble model with the SMOTETomek outperforms the base learners and improves the accuracy of diagnosis and prediction of the MACE occurrences in patients with ACS at early stage.