Rapid determination of free amino acids in Huangshan Maofeng tea using NIR spectra: Analysis and comparison of variable selection methods

Abstract

Tea is gaining global popularity appreciated for its valuable components and health benefits. This study investigated the non-destructive determination of free amino acids (FAA) content in Huangshan Maofeng tea using near-infrared (NIR) spectroscopy combined with chemometric methods. Firstly, preprocessing of the NIR signal was performed using the Savitzky–Golay method and Multiplicative Scatter Correction. Then, Competitive Adaptive Reweighted Sampling (CARS), Bootstrap Soft Shrinkage (BOSS), Iterative Variable Subset Optimization (IVSO) and Variable Combination Population Analysis (VCPA) were used to build support vector regression (SVR) models. Results in the current work showed that all four variable optimization algorithms enhanced the prediction accuracy of the models. Among them, the IVSO-SVR model performed the best. Its predictive determination coefficient ($R_p^2$) was 0.9615, and its root mean square error of prediction (RMSEP) was 0.0624. In summary, NIR spectroscopy combined with chemometric techniques provides a simple and efficient method for rapid, non-destructive quantitative detection of FAA content in tea, with potential for quality control applications.