Abstract:
Cardiovascular disease (CVD) continues to be the primary cause of global morbidity and mortality, highlighting the critical need for accurate risk prediction tools. Traditional risk models, such as the Framingham Risk Score, rely on cross-sectional data and conventional statistical techniques that often fail to capture complex interactions among risk factors. In this project, we present an ensemble machine learning framework designed to predict coronary heart disease (CHD) risk using the UCI Heart Disease dataset. Our approach incorporates advanced feature engineering techniques—such as generating interaction terms (e.g., age multiplied by systolic blood pressure), calculating differences (e.g., the gap between systolic and diastolic blood pressure), and applying logarithmic transformations—to better model non-linear relationships among variables. Individual models, including Random Forest, XGBoost, Gradient Boosting, and a Neural Network, are meticulously tuned via cross-validation and combined using stacking ensemble methods. Techniques like SMOTE are employed to address class imbalance, while explainable AI methods, particularly SHAP, provide both visual and textual insights into the contributions of key features. Performance is evaluated through ROC AUC, accuracy, precision, recall, and F1 score. This work aims to deliver a robust and interpretable CHD risk prediction tool that can enhance clinical decision-making and contribute to advancements in cardiovascular risk assessment.
