Machine-Learning Discrimination and Prediction of Flavor In Qingxiang-Xing Baijiu Fermented with Kiwi Fruit Residue

Abstract

A new solid-state-fermented Baijiu was developed to improve the comprehensive utilization efficiency of kiwifruit residues during fermentation. In this study, sorghum and kiwifruit residues were used as raw materials to brew a new type of Baijiu based on process of Qingxiang-xing Baijiu. The characteristics of volatile organic compounds (VOCs) were analyzed using sensory evaluation, gas chromatography-flame ionization detection (GC-FID), and headspace solid-phase microextraction coupled with gas chromatography-mass spectrometry （HS-SPME-GC-MS）. The results showed that the sorghum and kiwifruit residue mixed fermentation Baijiu (XQ-G) had the flavor of Xiaoqu Qingxiang-xing Baijiu (XQ) and a pleasant kiwifruit aroma. The total acid content of XQ-G was similar to that of XQ, while the content of most VOCs was significantly higher than that of the other three samples. The variable importance in projection of the VOCs detected using HS-SPME-GC-MS indicated that 26 different substances had a variable importance in projection greater than 1. Finally, machine-learning classification and prediction models were used for evaluation, and the best model for XQ-G was selected. This study clarifies the flavor characteristics of fruit and grain mixed fermentation in Baijiu and provides a reference for developing new types of Baijiu and the utilization of fruit residue.

Keywords: volatile organic compounds, Characteristic components, Linear regression and random forest, Machine-learning classification, Prediction models