Multivariate Classification of Geographical Origin of Herbal Medicines Using Terahertz Pulsed Absorption Spectroscopy

Abstract

Multivariate classification models of the geographical origin of herbal medicines are developed based on terahertz time-domain absorption spectroscopy. Four methods of cluster analysis are considered: hierarchical cluster analysis, k-means, k-nearest neighbors, and classification trees. Twelve spectral variables in the frequency range 0.1–0.6 THz have been selected for classification of five geographical locations of Gastrodia roots in decreasing order of their standard deviation in the explored set and transformed to a six-dimensional principal component space with the Mahalanobis metric. Classification of the considered herbal medicines by two nearest neighbors in this space is characterized by an accuracy and precision of 0.98, and a sensitivity of 0.982.