Healthcare Analytics for Early Coronary Heart Disease Prediction Using Machine Learning

Abstract

The Coronary Heart Disease (CHD) is among the foremost causes of global mortality, necessitating accurate and timely risk assessment for effective clinical intervention. Conventional predictive methodologies are constrained by their reliance on linear assumptions and a narrow set of variables, rendering them inadequate for capturing the multifactorial complexity inherent in CHD pathogenesis. This study presents a next-generation healthcare analytics framework that integrates three complementary predictive models—Logistic Regression, Decision Tree, and Graph-Based Neural Network (GNN)—to enable early and reliable CHD risk stratification.
The proposed framework processes patient clinical and demographic data through a structured pipeline encompassing preprocessing, feature engineering, graph construction, model training, and performance evaluation. The GNN leverages graph-structured representations of patient attribute relationships, enabling the detection of hidden interaction patterns that elude conventional models. Experimental evaluation on benchmark CHD datasets demonstrates that the GNN achieves a classification accuracy of approximately 84%, outperforming Logistic Regression (71%) and Decision Tree (62%). These results validate the superiority of relational learning in medical data analysis and underscore the framework's potential as a scalable clinical decision support tool