Accurate identification of kidney stone compositions is critical for optimizing clinical diagnoses and formulating individualized treatment plans. This study developed a rapid, noninvasive method for classifying kidney stone types via laser-induced fluorescence (LIF) spectroscopy combined with machine learning, establishing a 405 nm laser-based LIF system to collect spectral data for four common stone types: calcium oxalate monohydrate, anhydrous uric acid, magnesium ammonium phosphate hexahydrate, and carbonate apatite. Spectral data were preprocessed using standard normal variate transformation and normalization to reduce noise and morphological variability. Following feature selection and scaling, four classification models  (e.g., K-nearest neighbors, support vector classifier, random forest, and eXtreme gradient boosting) were constructed, with hyperparameters optimized via a Bayesian algorithm. All models performed well on an independent test set, with the support vector classifier achieving the highest average accuracy of 92%. This proposed technique enables reliable and efficient identification of the intraoperative stone components and holds significant potential for enhancing clinical diagnostics, guiding personalized treatment, and supporting recurrence prevention.