Improving Cardiovascular Disease Detection: A Comparative Approach with Boosting Algorithms and Imputed Features

This paper is written as an approach to the prediction of cardiovascular diseases using a merged dataset from five sources, incorporating imputed “Major Vessels” and “Thallium” features to enhance predictive accuracy. Synthetic Minority over-sampling Technique (SMOTE) is applied to handle imbalance in classes, followed by eight feature selection methods (ANOVA, Chi-Squared, Random Forest, LASSO, RFE, etc.) to rank features based on selection frequency. Machine learning algorithms, including Logistic Regression, Random Forest, SVM, Gradient Boosting, and XGBoost, were trained and evaluated using various metrics. LightGBM achieved the best performance with a test accuracy of 94.61%, F1 Score of 0.9528, and ROC-AUC of 0.9747. Key features such as Thallium, ST slope, exercise induced angina, chest pain, major vessels and ST depression were consistently selected, highlighting their predictive importance. The study concludes that boosting algorithms are highly effective for heart disease classification, outperforming other models, particularly when more features are included.