Dual-Platform GC–MS and 1H NMR Metabolomics with Explainable Machine Learning for Discrimination of Six Chinese Specialty Teas

Reliable discrimination of Chinese specialty teas requires objective characterization of their complex chemical profiles. This study investigated six specialty teas using sensory evaluation, GC–MS, 1H NMR spectroscopy, multivariate statistical analysis, and explainable machine learning. The analyses identified 138 volatile and 70 non-volatile compounds, revealing distinct sensory characteristics and chemical profiles among tea groups. PCA and PLS-DA demonstrated clear group separation. Among four machine learning algorithms, random forest achieved 100% classification accuracy for both volatile and non-volatile datasets under internal cross-validation. Feature importance analysis identified five volatile markers, including (±)-trans-nerolidol, geraniol, and nonanal, and eight non-volatile markers, including glutamine, theanine, and 1,3-dimethylurate. SHAP analysis further revealed their contributions to sample classification. These findings highlight the potential of integrating complementary chemical fingerprinting with interpretable machine learning for specialty tea discrimination. However, independent batch validation and targeted compound confirmation are required to establish the broader applicability of these markers for tea authentication.