Enhancing Heart Disease Prediction through SMOTE-ENN Balancing and RFECV Feature Selection

Abstract

Heart disease is the leading cause of mortality worldwide, exerting a significant influence on the national economic burden and productivity. The identification of heart disease is imperative for the prevention of more severe conditions, as it facilitates the detection of risks and symptoms at an early stage. The development of disease prediction models using machine learning has been extensively researched; however, the field continues to encounter challenges, including uneven data distribution and the presence of large, complex datasets. The proposed solution to these issues is the optimization of the Random Forest algorithm through the integration of the Synthetic Minority Over-sampling Technique and Edited Nearest Neighbor (SMOTE-ENN) with Recursive Feature Elimination and Cross-Validation (RFECV). The objective of these methods is to address the issue of data imbalance and to reduce irrelevant features, thereby enhancing the performance of the prediction model. The combination of SMOTE-ENN and RFECV consistently produces higher recall up to 0.984 and an optimal F1 score of 0.938. These results suggest that combining SMOTE-ENN data balancing and RFECV feature selection methods improves the performance of Random Forest, making it a promising approach for enhancing prediction models.