An Ensemble Machine Learning Approach for Cardiovascular Disease Prediction Using Feature Selection and Gradient Boosting

Heart disease remains a major global health issue, affecting millions of individuals across the globe. Early detection and prediction are crucial for reducing mortality rates. This study leverages advanced machine learning techniques to anticipate heart disease by employing the Kaggle heart disease dataset, which contains records of 70,000 patients with t2 clinical features. An ensemble learning approach is proposed for prediction. A Sophisticated Feature selection method is conducted using the LightGBM classifier. To address class imbalance, SMOTE-Tomek resampling combines synthetic oversampling with noise reduction. Power transformation is also used to equalize the feature distribution. In this study, PyCaret is utilized to compare multiple classifiers using stratified to-fold cross-validation, and from all of these, the top three models are selected. Then, applying soft voting blending and stacking with a Random Forest meta-model approach for further optimization. Among these, choose the top-performing model and make it final. The proposed ensemble model has an accuracy of 78.98% and a ROC-AUC of 86.0%, indicating that it is useful for clinical heart disease prediction.