Surface-enhanced Raman scattering integrating with machine learning for green tea storage time identification.

Abstract

The rapid and accurate identification of complex samples still remains a great challenge, especially for those with similar compositions. In this work, we report an integration strategy consisting of surface-enhanced Raman scattering (SERS) and machine learning to discriminate complex and similar analytes, in this case green tea products with different storage times. Surface-functionalized Ag nanoparticles (NPs) were used as a SERS substrate to reveal the changes in the sensory components of green tea with variable storage time. Principal components analysis (PCA)-based support vector machine (SVM) classification was used to extract the key spectral features and identify green tea with different storage times. The results showed that such an integration strategy achieved high predictive accuracy on time tag discrimination for green tea. The multiclass SVM classifier successfully recognized green tea with different storage times at a prediction accuracy of 95.9%, sensitivity of 96.6%, and specificity of 98.8%. Therefore, this work illustrates that the SERS-based PCA-SVM platform might be a facile and reliable tool for the identification of complex matrices with subtle differentiations.