This study focused on training and evaluating multiple machine learning models to predict surgical failure after percutaneous nephrolithotomy(PCNL). The study involved a retrospective analysis of 287 patients who underwent PCNL between the period of January 2018 and January 22, and these 229 of these patients were placed into the training set while the remaining 58 were placed in the testing set. From these patients, the study evaluated a total of 41 candidate features, or clinical variables, which generally include information about demographics, urine lab results, blood lab parameters, kidney stone geometry and composition, and the degree of composition. Using six different feature selection approaches, the team narrowed down the features that were most clinically significant in predicting surgical failure and narrowed the 41 features down to 8 in order to reduce model complexity and overfitting risk: stone size, stone skin distance, number of stones, presence of the medial calyx stone,, presence of the upper calyx stone, degree of hydronephrosis, and red cell distribution width coefficient. The team found that Voting Classifier was the best predictor of surgical failure after PCNL, demonstrating an area under the curve of 0.839, an accuracy score of 84.5%, and an F1 score(essentially an score of precision) of 0.842. Traditional scoring systems currently used have limited predictive power due to their reliance on linear relationships. By using machine learning models like Voting Classifier, more accurate predictions can be made because they are able to identify complex, non-linear relationships in clinical data.