Smart evaluation of tea quality: machine learning-assisted SERS for white tea vintage authentication and grade classification

Traditional sensory evaluation methods for tea—relying on empirical experience, visual colorimetry, and subjective taste perception—suffer from irreproducible inter-rater variability and inability to quantify bioactive markers, thus failing to establish scientifically verifiable quality standards. To address this, we developed an intelligent platform integrating surface-enhanced Raman scattering with machine learning for rapid and quantitative white tea analysis, enabling detection within just 20 min. Utilizing silver nano-coral substrates for signal enhancement and a 785 nm portable Raman spectrometer for full-spectrum data acquisition, we constructed a hybrid machine learning architecture featuring Random Forest for vintage classification (96% accuracy) and Gradient Boosting Regressor for epigallocatechin gallate (EGCG) quantification (R 2  = 0.98, the values of Root Mean Square Error, Mean Absolute Error, and Mean Square Error all approach zero). This approach established a definitive EGCG-content grading hierarchy: Yinzhen > Mudan > Gongmei > Shoumei, which is highly consistent with traditional sensory evaluation methods, while provides additional quantitative ability.