Rapid determination and variety identification of active ingredient contents in Chrysanthemi Flos by near-infrared spectroscopy and machine learning

In this study, a qualitative discrimination and quantitative prediction model for the rapid and effective identification of Chrysanthemi Flos (CF) was developed utilizing near-infrared spectroscopy (NIRS) in conjunction with ultra-performance liquid chromatography (UPLC) and multivariate algorithms. The contents of six active constituents-chlorogenic acid, luteolin-7-O-β-D-glucoside, 3,5-O-dicaffeoylquinic acid, tilianin, apigenin, and acacetin-in CF were quantified using UPLC. Significant variations were observed in the contents of these components across different CF varieties. Utilizing NIRS data, the subspace clustering algorithm effectively differentiated various CF varieties, achieving an accuracy rate of 95.7 %. Subsequently, the acquired NIRS data underwent preprocessing and feature selection variables, partial least squares regression prediction models for the six active components were successfully established, with R² values exceeding 0.8 for both the training and prediction sets. In conclusion, the integration of NIRS with machine learning technology has been demonstrated to be a rapid, effective, and feasible approach for the identification of CF varieties.