Machine learning discrimination of quality grades of base liquor integrating GC-TOF/MS and GC-IMS data analysis: Case study of strong-flavor Chinese baijiu.

The flavor of base liquor is critical to the grading quality of Baijiu. This study focuses on the base liquor grades of five different strong-flavor Baijiu brands. Headspace Solid-phase Microextraction Gas Chromatography Time-of-flight Mass spectrometry (HS-SPME-GC-TOF/MS) and Headspace Gas Chromatography Ion Mobility Spectrometry (HS-GC-IMS) identified 313 and 188 compounds respectively. 12 compounds correlated with quality grade (|ρ| > 0.5), including ethyl 2-methylbutanoate, ethyl isopentanoate, ethyl propanoate. Distinct compounds significantly related to the quality grades were identified. A fused dataset was combined with eight supervised machine learning. Random Forest (RF) and Logistic Regression (LR) performed excellently, achieving accuracy rates of 0.913 and 0.8696, F1 scores of 0.9167 and 0.8681, respectively, with strong validation results in Receiver Operating Characteristic (ROC) curves and confusion matrices. This study demonstrates that combining aroma profile with machine learning enables accurate and objective grading of strong-flavor Baijiu, offering a robust tool for quality control in the industry.