Near-infrared spectroscopy and machine learning-based technique to predict quality-related parameters in instant tea

Abstract

AbstractThe traditional method for analyzing the content of instant tea has disadvantages such as complicated operation and being time-consuming. In this study, a method for the rapid determination of instant tea components by near-infrared (NIR) spectroscopy was established and optimized. The NIR spectra of 118 instant tea samples were used to evaluate the modeling and prediction performance of a combination of binary particle swarm optimization (BPSO) with support vector regression (SVR), BPSO with partial least squares (PLS), and SVR and PLS without BPSO. Under optimal conditions, Rp for moisture, caffeine, tea polyphenols, and tea polysaccharides were 0.9678, 0.9757, 0.7569, and 0.8185, respectively. The values of SEP were less than 0.9302, and absolute values of Bias were less than 0.3667. These findings indicate that machine learning can be used to optimize the detection model of instant tea components based on NIR methods to improve prediction accuracy.