An Integrated Machine Learning Framework for Cardiovascular Disease Risk Stratification and Early Detection An Integrated ML Framework for CVD Risk Stratification and Early Detection

Abstract

Heart and blood vessel disorders—collectively termed cardiovascular disease (CVD)—represent one of the foremost contributors to mortality across the globe. This study presents a predictive framework developed to facilitate early identification of CVD risk, leveraging the 2023 Behavioral Risk Factor Surveillance System (BRFSS) dataset, which comprises roughly 308,854 survey respondents. To enhance data quality, the Interquartile Range (IQR) technique was applied to eliminate extreme observations lying beyond the central 50% of the feature distributions. Class imbalance—where disease-positive samples are substantially outnumbered—was addressed using SMOTE, applied strictly to training partitions to prevent information leakage. The study benchmarked four learning algorithms: K-Nearest Neighbors, Decision Tree, Random Forest, and LightGBM, each subjected to cross-validation-based hyperparameter search. LightGBM emerged as the top performer, recording 95.7% classification accuracy alongside an AUC of 0.958. A soft-voting ensemble yielded 93.9% accuracy with an F1 score of 0.939, surpassing previously reported accuracies of 79%–91%.