Advancing Interpretable AI for Cardiovascular Risk Assessment: A Stacking Regression Approach in Clinical Data from Bangladesh

Abstract

Cardiovascular diseases (CVDs) are complex conditions affecting a large portion of the global population, and their early, accurate, and timely prediction remains a significant challenge. Conventional CVD risk assessment often relies on limited parameters and fails to capture the complex interactions among genetic, lifestyle, and environmental factors. Recent machine learning studies have improved predictive performance; however, they often rely on small or retrospective datasets, lack real-time or external validation, and offer limited interpretability for clinical use. This study introduces a novel stacking ensemble framework that integrates ridge regression, Theil-Sen regressor, and gradient boosting regressor. To our knowledge, this is the first application of a regression-based stacking approach for CVD risk prediction that embeds explainable artificial intelligence as a core component, a combination rarely explored in low-resource healthcare contexts. Using a real-world dataset of 1,529 patients from Jamalpur Medical College Hospital, Bangladesh, the proposed model achieved 96% predictive accuracy, outperforming most existing methods. The dataset itself represents a rare contribution, as most prior studies rely on UCI, Framingham, or other benchmark repositories rather than contemporary hospital data from underrepresented populations. Through SHapley Additive exPlanations analysis, our model identifies BMI, diabetes, and blood pressure as the most influential factors, aligning with established medical knowledge and providing clinically actionable insights. Unlike prior black-box models, our framework not only improves prediction accuracy but also delivers transparent explanations that enhance trust and support public health decision-making. This integration of accuracy, explainability, and context-specific clinical insight underscores the novelty and practical relevance of our approach for advancing interpretable AI in CVD prediction, particularly in resource-limited healthcare settings. Received: 1 October 2025 | Revised: 10 November 2025 | Accepted: 30 November 2025 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement Data sharing is not applicable to this article as no new data were created or analyzed in this study. Author Contribution Statement Suhana Tasnim: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Project administration. Mohammad Mamun: Supervision. Safiul Haque Chowdhury: Writing – review & editing, Supervision. Mohammed Ibrahim Hussain: Supervision. Muhammad Minoar Hossain: Supervision.

Ähnliche Arbeiten