Fluorescence spectroscopy application for Argentinean yerba mate (Ilex paraguariensis) classification assessing first- and second-order data structure properties

Abstract

This paper uses fluorescence spectroscopy in order to evaluate the potential of this technique to classify Argentinean yerba mate from three commercial categories. An extraction method based on ultrasound for sample preparation is proposed. Acquired fluorescence first- and second-order data were analyzed by different chemometric approaches. Aimed at enhancing data interpretability and to reduce redundant information among the retained variables and maximize their relationship regarding the sample, several variable selection algorithms were evaluated. Two three-way algorithms, PARAFAC and n-PLS, were applied to EEMs. Several two-way algorithms such as PCA, SPA, GA, PLS were used previous to model data gathered from emission spectra recorded at the 410 nm excitation and unfolded EEMs. Sample categorization was performed using LDA or QDA as classification techniques. The ability of the models to predict samples was evaluated based on sensitivity and specificity. First-order models reached 89.6% on sensitivity and 90.9% on specificity. Second-order models achieved up to 100% sensitivity and specificity demonstrating that the EEM matrix provides relevant additional information revealing a significant effect on the classification results.