Discrimination between wild-grown and cultivated Gastrodia elata by near-infrared spectroscopy and chemometrics

Gastrodia elata is a notable medicinal and edible plant. Wild-grown Gastrodia elata is rare in nature and expensive for its medicinal value, which is often the subject of fraudulent practices by replacing them with inexpensive cultivated ones. Developing a rapid and effective method for identifying wild-grown Gastrodia elata is of great importance. The feasibility of combining near-infrared (NIR) spectroscopy with chemometrics for discriminating between wild-grown and cultivated Gastrodia elata was explored. A total of 141 samples were collected. Principal component analysis (PCA) was used for preliminary analysis. The Relief algorithm was used to select informative variables. Three kind of algorithms, i.e., and partial least squares-discriminant analysis (PLS-DA), extreme learning machine (ELM), Adaboost.M1 with decision tree as base classifiers, were used to construct predictive models. The Adaboost.M1 model using the selected 180 variables and decision stumps as the base classifier achieved the best performance on the test set, total accuracy of about 88 %. It seems that the combination of NIR spectroscopy, relief and Adaboost.M1 is potential for discriminating between wild-grown and cultivated Gastrodia elata.