Botanical-origin discrimination of Chinese monofloral honeys by untargeted metabolomics combined with machine...

Botanical-origin discrimination of monofloral honey remains challenging because botanical source, geography, and harvest year jointly shape chemical fingerprints. Here, untargeted UPLC-HRMS metabolomics combined with machine learning was applied to discriminate 233 authentic Chinese honey samples representing 11 monofloral types, together with 11 pure sugar syrup controls. After quality-controlled data processing, 11,752 features were detected, from which leakage-controlled least absolute shrinkage and selection operator (LASSO) retained 282 selected features. Among the evaluated classifiers, Random Forest performed best, achieving accuracies of 0.90 on the internal test set and 0.92 on an independent validation set of 63 market samples. t-distributed Stochastic Neighbor Embedding (t-SNE) visualized class-dependent patterns, and Random Forest feature importance and SHapley Additive exPlanations (SHAP) prioritized discriminative metabolomic features. This workflow provides a robust framework for botanical-origin discrimination of Chinese monofloral honeys, while targeted validation remains needed for compound confirmation and low-level syrup adulteration detection.