Rapid and simultaneous quantitative and discriminative analyses of liquor quality parameters with machine learning-assisted batch Raman spectroscopy: Synergistic instrumental upgrade and chemometric optimization

This study investigates the rapid and simultaneous assessments of two key liquor quality parameters - alcohol content and overall sensory quality - utilizing a batch Raman spectroscopic system and chemometrics, with a commercial baijiu (Chinese liquor) as the subject. An in-house designed motorized 12-cuvette tray facilitated stable and efficient spectral acquisition from 34 production batches of standard baijiu, supplemented by alcohol content-adjusted and overall sensory-disqualified samples, to form two separate datasets for the optimization and evaluation of chemometric approaches based on multivariate analysis and machine learning. The combination of dimension reduction with principal component analysis (PCA) and support vector regression (SVR) with a nonlinear kernel showed superior performances for predicting alcohol content and identifying sensorydisqualified samples. Expanding the alcohol content range of the training set enhanced the quantification capacity of the PCA-SVR model and yielded a relatively accurate alcohol content prediction (+/- 0.15 % v/v) for the tested standard baijiu. The PCA-SVR model built for sensory quality grading achieved an average precision of 93% for identifying disqualified baijiu samples that compositionally resemble standard ones. Based on the synergistic instrumental and chemometric optimization, the proposed machine learning-assisted batch Raman spectroscopic system offers a rapid, reliable, and integrated quality control tool for liquor production.