Machine learning-based classification and prediction of typical Chinese green tea taste profiles

The taste of Chinese green tea is highly diverse. In this study, a combination of unsupervised and supervised learning methods was utilized to develop a model for classifying and predicting typical Chinese green tea taste. Three clustering methods were assessed based on quantitative descriptive analysis (QDA) results, with the Hierarchical-K means method chosen to classify 88 tea infusions into seven distinct taste types. Electronic tongue sensors, near-infrared spectroscopy, and metabolomics, along with the analysis of key chemical constituents, were applied to construct various datasets as model data. The performance of four multivariate statistical methods and six artificial intelligence algorithms was compared across the three datasets. Dataset 3, comprising chemical components, taste activity value (Tav), and their ratios, achieved the highest accuracy. The random forest (RF) model achieved the highest accuracy (0.98) and Kappa value (0.97) in predictions. The results indicate that key chemical components, Tav, and their relationships are more critical for classifying green tea taste. This study can provide a more accurate representation and prediction of typical Chinese tea taste profiles from a consumer standpoint. Significant variations in sensory attributes and chemical composition were observed among the identified taste categories, with the MU type displaying the lowest TavTC (total Tav of catechins)/TavTAA (total Tav of amino acids) ratio, indicating the strongest umami and sweetness characteristics. The findings of this study offer the potential for the development of personalized tea products, thereby contributing to an enhanced consumer experience.