DEVELOPMENT AND COMPARATIVE EVALUATION OF MACHINE LEARNING MODELS FOR PERSONALIZED CARDIOVASCULAR RISK PREDICTION USING CLINICAL FEATURES

Abstract

Cardiovascular disease remains a major global health challenge, and the development of reliable computational approaches for individualized disease classification may support improved clinical decision-making. In this study, machine learning algorithms were developed and comparatively evaluated for the classification of heart disease using routinely available demographic, physiological, and clinical characteristics. The Cleveland Heart Disease dataset obtained from the UCI Machine Learning Repository was used as the study dataset. Four predictive approaches—logistic regression, random forest, support vector machine, and gradient boosting—were implemented to examine their ability to distinguish individuals according to the presence or absence of heart disease. A systematic analytical framework was applied, including missing-value treatment, categorical-variable encoding, feature scaling where appropriate, model optimization, and cross-validation. The predictive performance of the models was assessed using multiple complementary evaluation metrics, including the area under the receiver operating characteristic curve, sensitivity, specificity, precision, F1-score, and calibration. In addition to overall predictive performance, feature-importance analysis was incorporated to identify the demographic and clinical characteristics that contributed most substantially to model predictions and to improve the interpretability of the analytical framework. The comparative evaluation was designed to investigate whether nonlinear machine learning algorithms could provide advantages over conventional logistic regression for cardiovascular disease classification. The findings also demonstrate the potential value of interpretable predictive models as supportive tools for individualized cardiovascular assessment, while emphasizing that algorithmic predictions should complement rather than replace clinical judgment. Because the selected dataset was relatively small and contained historical diagnostic information, the results should be interpreted as retrospective classification findings and should not be considered equivalent to prospective prediction of future cardiovascular events. Overall, this study provides a reproducible machine learning framework for comparing conventional and nonlinear predictive approaches and establishes a methodological foundation for future research using larger, contemporary, and externally validated datasets in personalized cardiovascular healthcare.