Improving predictive accuracy: A comparative study of SMOTE and machine learning on heart disease data

This study investigates the impact of employing the Synthetic Minority Over-sampling Technique (SMOTE) to address data imbalance on the efficacy of several machine learning algorithms in predicting heart disease. We utilized the Cleveland Heart Disease dataset supplied from the UCI Machine Learning Repository and implemented the SMOTE approach to equalize the distribution of classes. Subsequently, we evaluated the efficacy of five machine learning algorithms: Random Forest, Logistic Regression, Support Vector Machine (SVM), Decision Tree, and Naive Bayes. Our study shows that using SMOTE significantly improves the performance of Random Forest, Logistic Regression, and Decision Tree algorithms. The enhancement is seen by the increase in accuracy, precision, recall, and F1 scores. On the other hand, the performance of SVM fell when SMOTE was applied, suggesting that SVM is vulnerable to synthetic data. The performance of Naive Bayes demonstrated a mixture of favorable and unfavorable results, characterized by improved precision and recall, albeit with a slight decrease in accuracy. The findings emphasize the need of choosing appropriate algorithms and strategies for handling imbalanced data in healthcare applications, thereby enhancing the accuracy and reliability of predictions. The study emphasizes the effectiveness of SMOTE in building robust machine-learning models for the diagnosis of heart illness and improving patient outcomes.