Machine learning prediction of pregnancy outcomes after vitrified-warmed bovine embryo transfer using field-derived variables.

Pregnancy outcomes after vitrified-warmed bovine embryo transfer (ET) remain variable, reflecting the influence of multiple field-derived factors. This exploratory study applied machine learning (ML) to predict pregnancy outcomes following vitrified-warmed bovine ET and identify field-derived variables contributing to the model-predicted probability of pregnancy. A total of 236 ET records were analyzed using recipient physiological, ovarian, reproductive management, and environmental and procedural variables. Although conventional statistical analysis identified significant differences only in lactation status and corpus luteum (CL) cavity, all field-derived predictors were retained for ML modeling. Six ML models were trained and internally evaluated using six-fold stratified cross-validation. Random forest showed the most balanced predictive performance, with modest discrimination (ROC-AUC = 0.703). SHapley Additive exPlanations (SHAP) analysis of the random forest model identified the temperature-humidity index (THI), CL volume, CL cavity, and lactation status as the main contributors to the model-predicted probability of pregnancy. SHAP dependence plots identified ranges associated with positive predictions, including THI of 58-77, CL volume ≥ 4.85 cm3, dominant follicle volume ≥ 0.79 cm3, body condition score of 3.0-3.5, and packaging-to-ET interval of 1.5-2.5 h. An exploratory artificial intelligence-based clinical decision support system prototype was also implemented to demonstrate individualized prediction output. These findings suggest that interpretable ML complements conventional statistical analysis by integrating multiple field-derived variables, including those with relatively low individual contributions, to support individualized pregnancy prediction after vitrified-warmed bovine ET. However, further validation using larger and independent datasets is needed to support future clinical or field application.