In the world, cardiovascular diseases (CVDs) are still the number one killer, so it is crucial that we develop prediction models that are both accurate and easily understandable.In order to improve prediction performance, this research investigates the potential of adopting ensemble learning approaches instead of traditional machine learning methods. To be more precise, we used a 1,190-record, 11-feature publically available Kaggle dataset. Using preprocessing and oversampling, we were able to rectify the class imbalance. The classical models that were used as evaluation baselines included KNN, SVM, DT, RF. Afterwards, a variety of ensemble approaches were utilized, including hard and soft voting, adaBoost and XGBoost boosting, random forest bagging, and stacking with Logistic Regression as the meta-classifier. Stacking resulted in an accuracy of 94.88%, proving that ensemble strategies routinely outperform individual models. By proving the framework could manage imbalanced data and back the long-term tracking of patients with chronic diseases, more case studies proved the system’s clinical relevance.
Tabbassum et al. (Tue,) studied this question.