Regional differences in physical fitness profiles of preschool children in Chongqing China revealed by machine learning.

Urban-rural disparities in physical fitness during early childhood remain insufficiently characterized. This study aimed to identify physical fitness phenotypes among preschool children in urban and rural settings and evaluate their predictability using anthropometric and demographic features. A total of 1049 children aged 3-6 years were assessed using standardized fitness tests. K-means clustering identified latent fitness profiles separately in urban and rural groups. Logistic regression, random forest, and XGBoost models were trained to predict cluster membership using anthropometric and demographic variables. Two distinct fitness phenotypes were consistently identified, with Cluster 1 demonstrating superior strength, power, and coordination ([Formula: see text]). Supervised models achieved moderate predictive performance (accuracy ≈ 0.79-0.80; AUC ≈ 0.85-0.86). Model interpretation indicated that age and height were the most influential predictors, while body composition and demographic variables contributed less. Preschool children exhibit distinct, region-dependent fitness phenotypes. While these phenotypes are primarily defined by motor performance, their prediction based on anthropometric data alone is moderate, highlighting the importance of functional assessments for early fitness evaluation.