Construction and Comparison of Coronary Heart Disease Risk Prediction Models Based on Lasso-Logistic Regression and Random Forest Models

Abstract

Coronary heart disease (CHD) is a prominent global cause of death, including genetic and environmental factors. High CHD mortality rates are observed in South Africa and China, and its prevalence is increasing among younger demographics due to lifestyle changes. Topic This study investigates the relationship between CHD and its risk factors using cardiovascular data from Framingham, Massachusetts, to develop a predictive model. Methods The study processed data using R 4.4.1 software, dividing it into training (70%) and testing (30%) sets. Feature variables for the model were selected through univariate analysis and Lasso regression. Prediction models were constructed using multifactorial Logistic regression and random forest machine learning algorithms. Model performance was assessed using ROC curves and confusion matrices. Results Lasso regression and univariate Logistic regression analysis, as well as stepwise Logistic regression analysis, indicated that gender, age, daily smoking, stroke history, hypertension, cholesterol, systolic pressure, and glucose are significant CHD predictors. The multifactorial Logistic regression model achieved an 85.6% classification accuracy, 85.8% precision, 99.5% recall, 0.921 F1 Score, and 0.717 AUC. The random forest model showed slightly lower performance with 85.2% accuracy, 34.8% precision, 99.2% recall, 0.919 F1 Score, and 0.615 AUC. Conclusion This study not only screened the influencing factors of coronary heart disease but also found that the CHD risk prediction model constructed by machine learning algorithm has good predictive performance, especially the Logistic regression model performed better on the data set of this study.