Predicting metabolic syndrome in Iranian adults using artificial neural networks: a comparative analysis with logistic regression based on demographic, lifestyle, and clinical indicators

Metabolic syndrome (MetS) is a major risk factor for cardiovascular diseases and type 2 diabetes, imposing a substantial economic and public health burden on Iran’s healthcare system. This study aimed to classify the risk of MetS in Iranian adults using an artificial neural network (ANN) and to compare its performance with that of logistic regression based on demographic, lifestyle, and clinical indicators. Data for this cross-sectional study were derived from the Fasa Cohort, part of the PERSIAN Cohort, and included 10,004 adults aged 35–70 years. Metabolic syndrome was diagnosed according to the modified ATP III criteria, and input variables were selected through a scoping review. An artificial neural network model based on a multilayer perceptron, along with multivariate and hierarchical logistic regression models, was implemented using SPSS version 27. Model performance was evaluated on the test set using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy. The prevalence of MetS was 21.4%, with a higher rate observed among women (30.4%). The triglyceride–glucose (TyG) index, visceral adiposity index (VAI), and waist circumference emerged as the strongest independent predictors. Both models demonstrated comparable discriminative performance, with AUC values ranging from 0.907 to 0.909. The ANN exhibited higher sensitivity in identifying high-risk individuals (68.3% vs. 54.3%) but lower specificity (90.9% vs. 94.0%), resulting in similar overall accuracy (85.9% vs. 85.4%).