Abstract: This study proposed a novel strategy for the rapid identification of nongxiangxing base Baijiu from different regions in Sichuan (Luzhou, Yibin, Chengdu, and Deyang) based on the fused data of gas chromatography (GC) and near infrared spectroscopy (NIR). A total of 225 base Baijiu samples were collected, and two classification models with accuracies of 75.1% and 79.2% were developed using random forest (RF) based on the flavor and spectral data, respectively. The models were found to have limitations in distinguishing complex geographical origins. Subsequently, a fusion classification model with an accuracy of 91.0% was constructed using the XGBoost algorithm based on the feature-level fusion of flavor and spectral information. The fusion model exhibited excellent robustness. The area under the receiver operating characteristic curve for Chengdu, Luzhou, and Yibin was above 0.95 each. Furthermore, by comparing the feature selection results of the fusion model with those of the flavor-based model, eight flavor markers to discriminate base Baijiu samples from different regions were identified through model consensus: ethyl acetate, ethyl lactate, ethyl heptanoate, ethyl formate, isobutanol, acetic acid, butyric acid, and propionic acid. This study not only provides insights for the development of geographical origin discrimination methods for nongxiangxing base Baijiu, but also offers theoretical support for elucidating the flavor chemistry of nongxiangxing base Baijiu from different regions of Sichuan.

Key words: nongxiang base Baijiu; flavor analysis; near infrared spectroscopy; machine learning; data fusion