CardioAI-Risk: An Interpretable and Fair Machine Learning Pipeline for Personalized Cardiovascular Risk Stratification

Abstract

Cardiovascular disease continues to be the leading cause of death globally, yet current predictive models often fall short in offering both individualized explanations and equitable performance across diverse populations. This study introduces CardioAI-Risk, an interpretable and fairness-conscious machine learning framework tailored for clinical application in preventative cardiology. Utilizing a logistic regression model guided by SHAP values, it incorporates ten key clinical and behavioral features such as blood pressure, cholesterol, age, gender, and lifestyle habits to assess patient risk. A refined risk threshold strategy enhances high-risk detection, while group fairness is systematically assessed using demographic parity and equalized odds metrics. The model attained an AUC of 0.87, with smoking, systolic pressure, and age identified as primary contributors to risk. Evaluation across demographic segments showed minimal performance gaps (∆parity <0.015), reinforcing its ethical validity. CardioAI-Risk offers clinicians a reliable, interpretable, and demographically balanced tool for personalized cardiovascular risk assessment.

Ähnliche Arbeiten