Reducing bias in coronary heart disease prediction using Smote-ENN and PCA

Abstract

Coronary heart disease (CHD) is a major cardiovascular disorder that poses significant threats to global health and is increasingly affecting younger populations. Its treatment and prevention face challenges such as high costs, prolonged recovery periods, and limited efficacy of traditional methods. Additionally, the complexity of diagnostic indicators and the global shortage of medical professionals further complicate accurate diagnosis. This study employs machine learning techniques to analyze CHD-related pathogenic factors and proposes an efficient diagnostic and predictive framework. To address the data imbalance issue, SMOTE-ENN is utilized, and five machine learning algorithms—Decision Trees, KNN, SVM, XGBoost, and Random Forest—are applied for classification tasks. Principal Component Analysis (PCA) and Grid Search are used to optimize the models, with evaluation metrics including accuracy, precision, recall, F1-score, and AUC. According to the random forest model’s optimization experiment, the initial unbalanced data’s accuracy was 85.26%, and the F1-score was 12.58%. The accuracy increased to 92.16% and the F1-score reached 93.85% after using SMOTE-ENN for data balancing, which is an increase of 6.90% and 81.27%, respectively; the model accuracy increased to 97.91% and the F1-score increased to 97.88% after adding PCA feature dimensionality reduction processing, which is an increase of 5.75% and 4.03%, respectively, compared with the SMOTE-ENN stage. This indicates that combining data balancing and feature dimensionality reduction techniques significantly improves model accuracy and makes the random forest model the best model. This study provides an efficient diagnostic tool for CHD, alleviates the challenges posed by limited medical resources, and offers a scientific foundation for precise prevention and intervention strategies.