Multicriteria decision making-based approach to classify loose-leaf teas

Abstract

Near infrared spectra of 75 different loose-leaf teas were analyzed based on their oxidational state: white, green, matcha, oolong, black, dark and pu-erh. Different spectral transformations (MSC, SNV and derivatives) and seven supervised linear and non-linear chemometric methods were performed. Classification methods were ranked based on their model performance metrics. In the ranking of the models, multicriteria decision making (MCDM) methods have crucial role, of which sum of ranking differences (SRD) method was used. SNV preprocessing showed better performance compared to MSC and FD + SNV. Among the models, linear support vector machine (lSVM) gave satisfactory performance regardless of the preprocessing. lSVM used on SNV preprocessed data proved to be the far best model, with 83.3 % accuracy. However, it is important to note that there are no general rules regarding model performances and proper testing is always advised. For such, multicriteria decision making models (and especially SRD) is strongly advised.

Keywords

• Sum of ranking differences
• Tea
• Quality
• NIR
• Machine learning