Machine learning based age-authentication assisted by chemo-kinetics: Case study of strong-flavor Chinese Baijiu

Abstract

The aged Chinese liquor, Baijiu, is highly valued for its superior organoleptic qualities. However, since age-authentication method and aging-mechanism elucidation of Baijiu is still in the exploratory stage, high-quality aged Baijiu is often replaced by lower-quality, less-aged product with fraudulent mislabeling. Authentic high-quality strong-flavor Baijiu was analyzed by gas chromatography-mass spectrometry. Total esters decreased with aging, while acids, alcohols, aldehydes, ketones, terpenes, pyrazines increased. Although concentrations of partial compounds showed non-monotonic profiling during aging, a close positive linear correlation (R² = 0.7012) of Baijiu Evenness index (0.55-0.59) with aging time was observed, indicating a more balanced composition in aged Baijiu. The reaction quotient (Qc) of each esterification, calculated by the corresponding reactant and product concentration, approached to the corresponding thermodynamic equilibrium constant Kc. This result demonstrated that the spontaneous transformation driven by thermodynamics explained part of the aging compositional profiling. Furthermore, an aging-related feature selection and an age-authentication method were established based on three models combined with five ranking algorithms. Forty-one key features, including thirty-six compound concentrations, four esterification Qc values and the Evenness index were selected out. The age-authentication based on neural network using forty-one input features accurately predicted the age group of Baijiu samples (F1 = 100 %). These findings have deepened understanding of the Baijiu aging mechanism and provided a novel, effective approach for age-authentication of Baijiu and other liquors.

Keywords: Baijiu age-authentication; Chemo-kinetics; Chemometrics; Composition evenness; Esterification; Feature selection; Machine learning; Reaction quotient.