One-class modeling for verification of botanical identity: a review

One-class modeling is a supervised multivariate botanical identification method based on principal component analysis (PCA) that constructs a model based only on the characteristics of the reference samples and uses the Q statistic as a combined metric. Test samples are judged to be similar (authentic) if their combined metric falls within the model limits or different (adulterated or contaminated) if the metric falls outside the model limits. This review initially considers three major factors affecting identification: the number of variables (univariate versus multivariate), the number of classes (one-class versus multi-class), and the type of analysis (quantitative versus qualitative). Multivariate analysis is commonly used for identification, providing a broader coverage of the identity specifications of the samples. With a combined metric, multivariate methods are analogous to univariate methods. One-class modeling and multi-class modeling employ different approaches for identification with one-class modeling being more flexible. While most methods to date have had a quantitative basis, qualitative methods are possible. This review focuses on multivariate, one-class modeling based on PCA. Examples are presented for the application of one-class modeling to identification of American ginseng (Panax quinquefolius), Echinacea purpurea, Black Cohosh (Actaea racemosa), and Maca (Lepidium meyenii). These examples demonstrate the utility and flexibility of one-class modeling.