Enhancing Heart Disease Prediction through Cross-Domain Transfer Learning from Related Health Conditions

Abstract

Heart disease remains a predominant cause of mortality worldwide, and the complexities in its predictive modeling are heightened by limited, domain-specific datasets. Traditional machine learning approaches often struggle to generalize across diverse populations due to data scarcity and the heterogeneous risk factors associated with heart disease. To address these limitations, this study explores a cross-domain transfer learning framework, which leverages knowledge from related health conditions, including diabetes, hypertension, and chronic respiratory diseases. This framework applies pre-trained models from these related domains to enhance prediction accuracy for heart disease in data-constrained environments. By adapting models trained on large datasets from overlapping medical domains, the proposed approach enriches heart disease models, allowing them to capture intricate risk patterns that might otherwise be overlooked. Experimental findings highlight the improved performance of transfer learning models over traditional heart disease models, particularly in terms of accuracy, sensitivity, and specificity. This cross-domain transfer learning approach not only addresses the challenges of limited heart disease datasets but also enhances predictive robustness, underscoring its potential for real-world clinical applications.