Rapid identification of adulteration in American ginseng powder using near-infrared spectroscopy combined with machine learning.

Rapid and reliable detection of adulteration in American ginseng (Panax quinquefolius L.) is essential for ensuring product integrity and safeguarding consumer health, as the powder is frequently adulterated with low-cost herbal substitutes. This study proposes a novel analytical framework integrating near-infrared (NIR) spectroscopy with machine learning modeling to identify adulteration involving Panax quinquefolius L, Angelica dahurica, Astragalus membranaceus, Panax notoginseng, and wheat flour. A total of 1680 samples were prepared, including 6 types of adulterants and 4 adulteration levels (20 %, 40 %, 60 %, and 80 %, w/w). Spectral data were preprocessed using noise reduction, baseline drift correction, and scatter correction. Principal component analysis (PCA) provided preliminary visualization of clustering trends, whereas four supervised classifiers-k-nearest neighbor (KNN), decision tree (DT), random forest (RF), and kernel extreme learning machine (KELM)-were comparatively assessed for classification performance. Among them, KELM achieved the best performance with an overall accuracy of 0.9560, precision/recall of 0.9629, and a Kappa coefficient of 0.9551 on the testing set. This study establishes an integrated NIR-chemometric framework using a hybrid-KELM for rapid, non-destructive, and interpretable authentication of American ginseng powder, where SHAP-based wavelength analysis highlights key spectral regions and provides practical guidance for developing simplified, application-specific NIR devices and broader detection of economically motivated adulteration in herbal products.