Classification and prediction model of ginseng year based on visible-near infrared spectroscopy and machine learning

Ginseng has the capability to modulate the immune system and augment the human body’s resistance. As a Chinese medicinal herb, it possesses substantial nutritional value. Accurate identification of ginseng’s cultivation year is of considerable significance for assessing its efficacy and value, ensuring fair market transactions, and facilitating cultivation-related research. Spectral data acquired from ginseng samples of five distinct years (8, 10, 15, 20, and 24 years) underwent collection and extraction. The extracted data were utilized to construct a dataset, which subsequently underwent preprocessing. Through the application of model algorithms, including principal component analysis, support vector machines, partial least squares (PLS), and random forest (RF) classification, classification prediction models were successfully developed to achieve year-based classification and identification of these five ginseng categories. Different modeling methods were evaluated using the established ginseng spectral dataset. PLS discriminant analysis modeling achieved an accuracy of 94.12±2% through fivefold cross-validation, whereas RF modeling obtained an accuracy of 88.54±2% under the same validation procedure. These two methods are identified as the most effective approaches for classification prediction.