Portable near-infrared spectroscopy combined with machine learning algorithms for the origin identification and quality evaluation of Acanthopanax senticosus

Abstract

The aim of this study is to develop a comprehensive quality assessment method for Acanthopanax senticosus using near-infrared (NIR) spectroscopy combined with machine learning algorithms. A qualitative discriminant analysis model was established to distinguish the samples of Acanthopanax from different habitats, and overall prediction accuracy was more than 85%. Moreover, a partial least squares regression (PLSR) model was built to predict the response values of Acanthopanax senticosus at each retention time, with the Acanthopanax senticosus dataset serving as the Y matrix and the NIR spectral dataset serving as the X matrix. The transformed Acanthopanax senticosus fingerprint closely matched the actual fingerprint, exhibiting a high average similarity between the predicted and measured fingerprints for test set samples (Pearson correlation coefficient = 0.8553, cosine similarity = 0.8627). The results of the cluster analysis demonstrated a high level of consistency. Converting NIR spectral fingerprints into HPLC fingerprints enhanced interpretability and more clearly displayed the components responsible for the diversity among samples from different origins. Meanwhile, the partial least squares regression (PLSR) model and support vector machine regression (SVR) model for the prediction of the contents of moisture, total solids, protocatechuic acid, syringin and chlorogenic acid contents were successfully established. The NIR spectra of Acanthopanax senticosus can accurately predict the values of moisture, total solids, protocatechuic acid, syringin and chlorogenic acid content, in which the RMSEP values for PLSR and SVR are 0.176, 0.812, 0.003, 0.026, and 0.007 and 0.182, 0.797, 0.003, 0.025, and 0.006, respectively. The predicted values of the five chemical compounds were 6.09–7.81, 4.8–9.5, 0.003–0.036, 0.066–0.233 and 0.015–0.055. Overall, the results indicate that this rapid quality evaluation system can be utilized for the quality control of Acanthopanax senticosus. The approaches used in this work can be used to identify the compound contents and provenance of herbal medicines and food for quality control measurements, potentially for online and field analyses.