MODELE D’APPRENTISSAGE PAR ENSEMBLE BASE SUR LE DOUBLE STACKING POUR LA PREDICTION DU SYNDROME METABOLIQUE

Abstract

Metabolic syndrome is a major public health issue because of its strong association with cardiovascular disease, stroke, and type 2 diabetes. This study proposes a predictive model based on a Double Stacking architecture using NHANES 1998-2018 data. The methodology includes preprocessing demographic, clinical, and biological variables, training several base classifiers, and combining them through two successive meta-learning levels. The results show that the proposed model improves predictive robustness and reaches an accuracy of 91.25%, an F1-score of 86.18%, and a ROC-AUC of 0.9687. These findings suggest that Double Stacking is a promising approach for the early prediction of metabolic syndrome.