Predicting Heart Disease Risk Using Machine Learning: A Comparative Analysis of Linear and Nonlinear Models

Abstract

Heart disease is one of the leading causes of mortality worldwide, and early risk prediction plays a vital role in reducing its impact. Traditional assessment methods such as the Framingham Risk Score are widely used but rely on linear assumptions, which can overlook complex interactions between clinical factors. Machine Learning (ML) offers promising alternatives by modeling these nonlinear relationships. In this study, the predictive capabilities of two interpretable machine learning models—Logistic Regression and Random Forest—are compared using a clinical dataset of 918 patient records. The dataset includes key features such as age, sex, cholesterol, resting blood pressure, and heart rate. The Random Forest model slightly outperforms Logistic Regression in terms of accuracy (90.2% vs. 88.6%) and AUC (93.5% vs. 92.9%), while both models achieve high recall (93.1%), which is critical in minimizing missed diagnoses. Feature importance analysis using SHAP values identifies MaxHR, ST_Slope, and cholesterol as key predictors. This study highlights the potential of accessible, interpretable ML methods to support clinical decision-making in cardiovascular care while ensuring transparency and reproducibility.