Comparative Evaluation of Machine Learning Models for Cardiovascular Disease Prediction

Abstract

Cardiovascular diseases (CVDs) persist as a critical global health burden, accounting for over 30% of annual mortality worldwide. While machine learning approaches show promise in early-stage risk identification, existing research predominantly focuses on isolated model validation without systematic comparative analysis. To address this gap, our study conducts a rigorous multi-algorithm evaluation, leveraging Logistic Regression (LR), Random Forest (RF), and XGBoost classifiers trained on clinically validated CVD datasets. Methodologically, this study implement stratified 5-fold cross-validation with adaptive class-weight balancing to mitigate data imbalance issues, while embedding recursive feature elimination for optimal predictor selection. Performance benchmarking across three critical dimensions-precision, recall (sensitivity), and ROC-AUC-reveals RF’s consistent superiority in both discriminative power and operational stability. Specifically, RF generates the most favorable sensitivity-specificity trade-off curve and attains significantly higher diagnostic sensitivity compared to LR and XGBoost, essential for minimizing false negatives in clinical screening scenarios. The validated model establishes a scalable risk stratification pipeline for community healthcare systems, enabling timely interventions while demonstrating methodological replicability for predictive epidemiology in resource-constrained primary care settings. This framework bridges a critical feasibility gap in translating algorithmic innovations into deployable clinical tools.