Machine Learning Approach of Polyphenol Content Prediction for Fresh Tea Leaves Using NIR Spectroscopy

Tea ranks as the most popular non-alcoholic beverage globally. The presence of polyphenolic compounds in tea contributes to its antioxidant properties, making tea a popular beverage among consumers. Colorimetric testing is commonly used to assess polyphenol content (FC value) in tea leaf using Folin Ciocalteu’s (FC) phenol agent. The FC values thus obtained for 12 tea leaf samples chemically are used as the reference for the present work. A spectrometer was employed to acquire the FT-NIR diffused reflectance spectral data from the powdered samples. Principle component regression (PCR) and partial least square regression (PLSR) are utilized in this paper to predict FC values. Prior to predicting the FC values, MSC, SNV, and Savizky–Golay filter-based first-order and second-order derivative techniques are used. For the experiment, twelve fresh tea leave samples are being considered. The total polyphenol content (FC values) is predicted using comparative research. The performance indices MSE, MAE, and correlation factor (R2 score) represent the efficacy of the preprocessing method and regression model combination.