Abstract
Introduction: Metabolic syndrome (MetS) is increasingly prevalent worldwide and under-addressed, with emerging interest in using mobile technology for health management. Effective interventions hinge on reliable prediction models. Objectives: This study aims to develop an algorithm to estimate MetS risk using only lifestyle factors and assess its impact on patient screening and quality of life enhancement. Methods: Utilizing data from the Korean National Health and Nutrition Examination Survey (2010–2018), we trained three non-invasive classifier models—artificial neural network (ANN), XGBoost, and LightGBM—for binary classification. We evaluated model performance using sensitivity, specificity, AUROC, and AUPRC metrics and explored quality-adjusted life years (QALYs) improvements. Results: Machine learning models demonstrated superiority over traditional logistic regression, with LightGBM achieving the highest AUROC and accuracy (AUROC 0.84; accuracy 0.74). Decision curve analysis highlighted significant differences in external datasets. MetS severity was strongly associated with QALYs (p < 0.0001), predicting substantial QALY gains across MetS categories. Conclusion: The developed model enhances MetS risk assessment accuracy and underscores the importance of incorporating gender-specific factors in predictive models.