A Novel Multi-Stage Stacked Learning Framework for Cardiovascular Risk Stratification

Abstract

Cardiovasculardiseases (CVDs) remain the leading cause of global mortality, 
necessitating accurate and early risk stratification to improve clinical outcomes. 
This paper presents a novel multi-stage stacked learning framework for robust 
cardiovascular risk prediction by leveraging heterogeneous machine learning 
models in a hierarchical architecture. The proposed framework consists of a 
feature extraction layer followed by multiple base learners, including Support 
Vector Machine (SVM), Random Forest, XGBoost, and deep learning-based 
models, which capture diverse statistical and nonlinear patterns from clinical and 
imaging data. A meta-classifier aggregates the predictions of these base learners 
to generate a unified and optimized risk score. Experimental evaluation 
conducted on a benchmark cardiovascular dataset demonstrates that the proposed 
stacked architecture outperforms conventional single-model approaches and 
traditional ensemble techniques in terms of accuracy, precision, recall, and F1
score. The results highlight the effectiveness of multi-stage stacked learning in 
enhancing predictive robustness and supporting reliable clinical decision-making 
for early cardiovascular risk stratification.