Preterm birth phenotype prediction models

Production machine-learning models for risk stratification of preterm birth by phenotype (spontaneous vs indicated), trained on a single-centre cohort of 13,311 pregnancies in 12,509 women (Petrozavodsk Regional Perinatal Centre, 2022–2025). Three models for the M2 prediction window (≤24⁺⁶ weeks of gestation; internal filename suffix `_m12` denotes the same level as article's M2, using features from both first and second trimesters): combined PTB, spontaneous phenotype, indicated phenotype. Stacking ensemble (XGBoost + LightGBM + CatBoost + RandomForest + meta-LogisticRegression); iatrogenic model is a regularised logistic regression. Includes preprocessing code, predictor wrappers, full feature list (185), cohort medians for imputation, and an example notebook with four synthetic clinical profiles.