A novel strategy for rapid quantification of multiple quality markers and authenticity identification based on near-infrared spectroscopy and machine learning algorithms, Fructus Gardeniae as a case study

Fructus Gardeniae (FG) is a commonly used medicinal and edible herb, with its quality influenced by factors such as harvest batches, geographical origins, and adulteration. Traditional methods for quality control of FG are time-consuming and invasive, creating an urgent need for a more rapid and efficient alternative. This study aims to develop a rapid and cost-effective strategy for FG quality control by utilizing near-infrared (NIR) spectroscopy combined with machine learning algorithms. Partial least squares (PLS) regression models were optimized through spectral preprocessing and variable selection techniques to predict four key quality markers in FG. In addition, four classification models, including orthogonal partial least squares discriminant analysis (OPLS-DA), random forest (RF), gray wolf optimizer-support vector machine (GWO-SVM), and convolutional neural network (CNN), were employed to distinguish between FG, Fructus Gardeniae Grandiflorae (FGG), and adulterated mixtures (MIX). The optimized PLS models successfully predicted the contents of genipin-1- β -D-gentiobioside, geniposide, crocin I, and crocin II, with residual predictive deviation (RPD) values of 2.4870, 3.1134, 1.4775, and 1.5316, respectively. The classification accuracy for OPLS-DA, RF, GWO-SVM, and CNN models was 82.93%, 82.93%, 95.12%, and 92.68%, respectively, with GWO-SVM achieving the best performance. Only two of the 41 validation samples were misclassified by the GWO-SVM model. In conclusion, this study provides a comprehensive and rapid solution for detecting FG adulteration and evaluating quality, offering significant improvements in efficiency compared to traditional methods.