The study developed machine learning models to predict recurrent extrahepatic bile duct stones after common bile duct exploration. LASSO regression selected 8 predictors and created 9 models that were evaluated using AUC, accuracy and other metrics. The best results were achieved by the random forest (RF) model with an AUC of 97.99% in training, 93.66% in verification and 83.1% in the external cohort, with an accuracy of 0.953/0.902/0.829. The SHAP analysis showed that the most significant risk factors are the maximum diameter of the stone, the diameter of the common bile duct and direct bilirubin. These factors have non-linear effects, such as increased risk for stones larger than 15 mm, and exhibit synergistic interactions. The RF model outperforms others in generalizability and enables personalized risk assessment to reduce postoperative recurrence.