Rapid authentication of Cuscutae Semen with imbalanced sample categories using a handheld near infrared spectrometer

Chinese herbal medicines, primarily derived from plants and natural sources, are widely incorporated into the formulation of health foods and dietary supplements. Ensuring their authenticity is crucial for maintaining therapeutic efficacy. This study introduces a method for rapid authentication of Chinese herbal medicines using a handheld near-infrared spectrometer coupled with chemometrics. Focusing on Cuscutae Semen, prone to market adulteration, the method involves spectral data collection, data preprocessing, feature processing, and classification algorithm. To address the challenge of imbalanced datasets prevalent in practice, synthetic minority over-sampling technique with tomek links (SMOTETomek) comprehensive data sampling was applied, enhancing model discrimination. The resulting model, combining Savitzky-Golay smoothing with the first derivative and a random forest classifier (SGFD_RF), achieved high accuracy in category authentication, with macro-averaged area under the curve (AUC_macro) scores of 0.997 (cross-validation) and 0.945 (test set). And the f-score and recall of the test set reached 0.954 and 0.955, respectively. For content authenticity detection, the SGFD_RF model displayed outstanding performance, with AUCs of 0.995 (cross-validation) and 1.000 (test set). Both f-score and recall of the test set reached 1.000. The study also demonstrated that the competitive adaptive reweighted sampling algorithm could reduce data dimensionality and training time, while providing even more precise classification with only 8 features. This approach offers a rapid and reliable solution for on-site herbal medicine authentication.