Machine learning-driven multidimensional tea profiling from a single SERS spectrum: toward practical application

Multidimensional tea profiling is crucial for consumer experience, brand reputation, and market integrity. Here, we introduce a rapid and systematic strategy for multidimensional tea authentication based on a single surface-enhanced Raman spectroscopy (SERS) measurement integrated with machine learning. We first demonstrate, through machine learning analyses of individual variables, that a single SERS spectrum inherently encodes multidimensional information encompassing tea category, grade, storage quality, and pesticide residue levels. Partial least squares discriminant analysis enabled highly accurate classification of tea categories and grades (accuracies ≥ 98.9%), while support vector machine regression achieved precise prediction of storage quality and pesticide residue levels (R² > 0.99). Building upon these results, we developed a two-tier framework that integrates these multidimensional predictions from a single spectrum, enabling comprehensive authentication in one step. This framework was further implemented in a user-friendly application for real-time, single-spectrum analysis, achieving an overall accuracy of 98.2%. Collectively, this work demonstrates that SERS coupled with machine learning provides an efficient, cost-effective, and scalable approach for multidimensional tea authentication, offering a foundation for collaborative improvement and data sharing across the tea industry.