Early Prediction of Cardiac Arrest

Abstract

Cardiovascular diseases (CVDs) are still the most prevalent cause of death globally. Early identification is key to enhancing treatment outcomes and minimizing death rates. Although earlier studies had shown the applicability of machine learning (ML) models like SVM, logistic regression, and decision trees, this paper offers a novel method that uses sophisticated ensemble methods, viz., Random Forest and XGBoost, combined with feature engineering and interpretability. Employing the Cleveland and Statlog heart disease datasets, we perform thorough preprocessing, feature inspection, and model optimization via hyperparameter tuning and cross-validation. We evaluate on the usual metrics (Accuracy, Precision, Recall, F1-score, AUC) and use SHAP for interpretability. The presented XGBoost model obtains better performance with an accuracy of 93.45% and high interpretability, making it a suitable decision-support tool in hospitals.