The study developed and internally validated machine learning models to predict admission hypothermia (axillary temperature <36.5°C) in preterm infants (<37 weeks' gestation) at a tertiary center in China from January 2017 to January 2025. The retrospective cohort was divided into a training (70%) and validation (30%) set, predictors selected by LASSO (11 factors: gestational age, birth weight, ambient temperature, transport time, congenital condition, preheated incubator and others). Six models were developed: logistic regression, decision tree, random forest, support vector machine, artificial neural network, and naive Bayes. In the validation cohort, they achieved AUCs from 0.78 to 0.86, besting logistic regression and artificial neural network (AUC=0.86). Logistic regression had good calibration and interpretability, SHAP analysis showed lower gestational age, lower birth weight, lower ambient temperature and longer transport time as the main risk factors. The models use commonly available perinatal and environmental variables to effectively predict hypothermia.