Objective To compare the performance of five machine learning models with the traditional logistic regression model in predicting the risk of prolonged intensive care unit (ICU) stay after cardiac valve replacement, and to construct an interpretable predictive model to provide decision support for the early clinical identification of high-risk patients. Methods A retrospective analysis was conducted on the clinical data of patients who underwent cardiac valve replacement at the First Affiliated Hospital of Anhui Medical University from August 2021 to April 2025. Five machine learning models, including support vector machine (SVM), random forest (RF), naive Bayes, decision tree (DT), and extreme gradient boosting (XGBoost), were constructed to predict the risk of prolonged ICU stay (≥3 d), with the traditional logistic regression model included as a control. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, F1-score, and Brier score. The SHapley Additive exPlanations (SHAP) method was applied for the model interpretability analysis. Results A total of 621 patients who underwent cardiac valve replacement were included, comprising 319 (51.4%) males with a median age of 60 (interquartile range 54-68) years; 182 (29.3%) patients experienced a prolonged ICU stay. Among all models, the RF model exhibited the best overall performance, with an AUC of 0.864, accuracy of 0.770, sensitivity of 0.797, specificity of 0.756, F1-score of 0.703, and a Brier score of 0.149. The final optimized model incorporated 13 important variables: operative time, intraoperative lactate, left ventricular ejection fraction, hemoglobin, left atrial diameter, use of anticoagulants, blood urea nitrogen, cryoprecipitate transfusion, history of cerebrovascular disease, New York Heart Association (NYHA) cardiac function class≥Ⅲ, use of calcium channel blockers, history of prior cardiac surgery, and hepatic dysfunction. Conclusion The RF model possesses excellent predictive performance and interpretability in predicting prolonged ICU stay after cardiac valve replacement, demonstrating superior overall capability. It can serve as a valuable reference for the early clinical identification of high-risk patients and the optimization of intensive care medical resource allocation.