Objective To establish and validate interpretable machine-learning (ML) models for early assessment and identification of Chinese women at risk of excessive gestational weight gain (EGWG). Methods We performed a prospective observational study with pregnant women whose gestational age averaged 19 weeks or less. These women attended the obstetric clinic of a tertiary hospital in CentralSouth China between January and June 2023, and again from April to May 2024. Women completed standardized questionnaires, and their gestational weight gain (GWG) was recorded until delivery. We conducted feature selection by applying the Boruta algorithm together with the least absolute shrinkage and selection operator (LASSO) algorithm, We used four ML models—the logistic regression(LR), light gradient boosting machine(LightGBM), extreme gradient boosting(XGBoost), and random forest (RF) models—optimizing its hyperparameters by grid search and 5-fold crossvalidation. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), confusion matrix, kappa statistic, calibration curves, and decision curve analysis. To enhance model interpretability, the SHapley Additive exPlanations (SHAP) framework was applied to quantify and rank the contribution of individual predictors. Based on the key predictive features, a web-based interactive calculator was developed using Python and the Flask micro-framework to facilitate clinical application. Results We enrolled 578 pregnant women in all. The combined use of the Boruta and LASSO algorithms screened ten critical predictors. The LightGBM model showed superior predictive performance with an accuracy of 88.6%, sensitivity of 87.5%, specificity of 89.9%, kappa statistic of 0.770, and AUC of 0.926 (95% CI: 0.889–0.962) in the test cohort. The SHAP analysis indicated that the body image in pregnancy, protective motivation for gestational weight management, parity, the weekly frequency of consuming sugar-sweetened beverages, desserts, and Western-style fast food, and moderate-intensity physical activity time were the major determinants that influenced model prediction. An online calculator was developed and made available for clinicians at: http://39.103.64.176/ . Conclusions We established an interpretable ML model for predicting the risk of EGWG. The LGBM model exhibited higher predictive accuracy and may serves as a powerful tool for the early detection and individualized management of the EGWG risk among Chinese pregnant women.
He et al. (Mon,) studied this question.