SmartHeart: A conceptual framework for explainable machine learning in cardiovascular risk prediction

Cardiovascular diseases (CVDs) remain the leading cause of mortality worldwide. Early prediction and timely intervention are critical to reducing the burden of heart disease. This study proposes SmartHeart, a conceptual framework that integrates structured clinical data with a proposed real-time data acquisition pipeline for interpretable cardiovascular risk prediction. A publicly available heart disease dataset - aggregated from multiple clinical sources and shared in a merged, cleaned form on Kaggle, containing 11 clinical variables and 1190 patient records, was used to train and evaluate six supervised machine learning models: Support Vector Classifier (SVC), Random Forest, XGBoost, CatBoost, AdaBoost, and Extra Trees Classifier. Following rigorous preprocessing, model performance was assessed using a stratified nested 5-fold cross-validation framework, where an inner loop optimized hyperparameters and an outer loop provided robust internal performance estimation, followed by final evaluation on an independent held-out test set. Among all models, Random Forest achieved the highest performance, with an accuracy of 92.86 % and an AUC of 97.14 %, supported by 95 % confidence intervals and pairwise t-tests confirming its statistical superiority. To enhance interpretability, SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) were applied to explain individual predictions, identifying features such as chest pain type, ST slope, and maximum heart rate as key contributors. While the real-time component remains at the architectural and conceptual stage, the proposed SmartHeart framework lays the foundation for future integration into cloud-based healthcare systems, enabling explainable and proactive cardiovascular risk assessment.