Improving Heart Disease Severity Prediction Using SMOTE for Imbalanced Data

Abstract

The heart disease is a prevalent and potentially fatal condition affecting individuals worldwide. In this study, we address the challenge of predicting the severity of heart disease using supervised learning techniques. Leveraging a dataset comprising various demographic and clinical attributes, we propose a solution that employs machine learning models to accurately predict the severity level of heart disease. Among the evaluated models, Random Forest emerges as the top performer, showcasing exceptional precision, recall, accuracy, and F1-score across all severity levels, with an overall accuracy of 98.8%. This highlights the robustness of the Random Forest model in accurately classifying instances across different severity levels. Following closely behind, the KNN algorithm demonstrates commendable performance, achieving an accuracy of 92% and showcasing competitive precision, recall, and F1-score values, particularly for higher severity levels. Despite its notable aspects, XGBoost ranks third among the evaluated models, with an accuracy of 90.4%. While XGBoost excels in certain aspects, such as recall for Level 3 severity, it falls short in overall performance compared to Random Forest and KNN. For future research, exploring ensemble methods that combine the strengths of different algorithms could yield even better classification results, providing avenues for further improvement in predicting the severity of heart disease