To perform feature selection with detected Raman spectroscopic data and establish a pungent flavor discrimination model based on the detection and analysis of Raman spectroscopy.Methods:After sample preprocessing，132 pungent Chinese medicines and 156 non-pungent Chinese medicines were analyzed using a SEED 3000 Raman spectrometer.Raman spectra were obtained for each Chinese medicine and quantified at 1 cm-1 intervals.Feature selection was performed on the quantified Raman data using random forest(RF) and extreme gradient boosting(XGBoost) algorithms to identify Raman shifts and peak intensities closely related to pungent flavor.Discrimination models were then built using five algorithms，i.e.，RF，K-nearest neighbors(KNN)，gradient boosting machine(GBM)，naive Bayes(GNB)，and adaptive boosting(AdaBoost)，and the models were evaluated.Results:Pungent Chinese medicines exhibited high-intensity Raman scattering in the 2 500 to 3 000 cm-1 range compared to non-pungent ones.Feature selection based on the top 100 Raman shift data from RF and XGBoost algorithms identified the most important features.Among all models，the GBM algorithm demonstrated the best performance with an area under the curve(AUC) of 0.978，an accuracy of 0.943，and a precision of 0.970.Conclusion:Raman spectroscopy is significantly related to the pungent flavor of Chinese medicines and can be used to quantitatively characterize pungent flavor.Combining Raman spectroscopy with the GBM algorithm allows for efficient and accurate identification and analysis of pungent Chinese medicines.