From Data to Decision: Interpretable and Reliable Heart Disease Risk Prediction

Abstract

Background: Heart disease is the leading cause of morbidity and mortality worldwide, highlighting the need for accurate and clinically meaningful risk prediction tool. Methodology: A retrospective analysis was performed on a dataset of 918 patients with complete clinical information. The relationship between variables and disease presence were explored using correlation and dependence measure. To prediction was made using logistic regression, linear discriminant analysis, and generalized additive models. The performances were evaluated using cross-validation, discrimination metrics, and calibration analysis. Results: Functional and exercise-related variables, particularly Oldpeak, maximum heart rate, chest pain type, and ST slope showed the strongest correlation with disease presence. All models achieved AUROC values higher than 0.91. Logistic regression and generalized additive models performed similarly, with no statistically significant difference between them. Calibration analysis showed a good agreement between predicted and observed probabilities. Conclusion: Heart disease prediction can be achieved with high accuracy using clinically accessible variables and interpretable models. Our findings suggest that simpler approaches remain suitable for practical clinical applications, while preserving transparency and usability.