Research on Risk Prediction of Coronary Heart Disease Based on Machine Learning

Coronary Heart Disease (CHD) is one of the leading causes of death worldwide, and precise prediction of its ten-year incidence risk is crucial for early intervention. This study constructs a multi-dimensional risk prediction model integrating clinical indicators, lifestyle habits, and physiological features based on the Framingham Heart Study dataset (4,240 samples). Five medical interaction features are constructed to enhance model representation. Categorical variables are encoded via Stratified K-fold Target Encoding, and sample imbalance is addressed using the SMOTEENN algorithm. XGBoost is used for model training. The results show that the model achieved an accuracy of 98.2%, an Area Under the Curve (AUC) of 0.9986, and a recall rate of 99.1% for high-risk samples in the test set. The calibration curve indicated a high consistency between predicted probabilities and actual risks (Brier Score = 0.0130). This model provides an efficient decision-support tool for clinical early risk stratification and demonstrates promising application potential.