Development and Interpretation of a Three-Year Risk Prediction Model for Metabolic Syndrome Based on Imbalanced Data

Le résumé fourni par la source

Metabolic Syndrome (MetS) is a common precursor to cardiovascular disease, diabetes and other diseases, posing a growing threat to public health systems.To enable efficient and interpretable individual risk identification, this study developed a 3-year MetS prediction model, incorporating 31 features from demographics, behaviors, anthropometrics, and dietary intake. To address class imbalance, four strategies-SMOTE, NearMiss, SMOTEENN, and class weighting-were applied at both data and algorithm levels. LightGBM was adopted as the base classifier, and Bayesian Optimization was employed for parameter search. The best results were achieved under the SMOTEENN-BO-LightGBM model, with an AUC of 0.792 and recall of 0.812. Model interpretation via SHAP highlighted BMI, waist circumference, sex, age, and blood pressure as key risk factors. This study develops a MetS prediction model that integrates data balancing, hyperparameter optimization, and interpretable analysis, without relying on laboratory-based measurements, demonstrates practical feasibility for community-level deployment, and provides decision support for MetS screening and high-risk population intervention.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.