Rapid identification and semi-quantification of adulteration in walnut oil by using excitation–emission matrix fluorescence spectroscopy coupled with chemometrics and ensemble learning

Abstract

Walnut oil (WNO) is a high-quality vegetable oil with high nutritional value, but it may be the target for adulteration and mislabeling. Therefore, this work proposes the use of excitation–emission matrix fluorescence (EEMF) spectroscopy coupled with chemometrics and ensemble learning to identify and semi-quantify the adulterated WNO. Herein, a total of 711 vegetable oil samples were analyzed. A five-component alternating trilinear decomposition (ATLD) completed the EEMF spectra characterization of 426 pure oil samples and provided chemically meaningful information. Five authenticity identification methods, including eXtreme gradient boosting (XGB), gradient boosting decision tree (GBDT), random forest (RF), multi-way partial least squares discriminant analysis (N-PLS-DA), and k-nearest neighbor (k-NN), were used for the discrimination of WNO and counterfeited WNO. Besides, PLS regression (PLSR) models were established to semi-quantify the adulterated levels of WNO concentration in edible blended oils. The results showed that ensemble methods obtained better performance, with the correct classification rate (CCR) of cross-validation being ranged 82.7–100 %, and the value of RMSEP is less than 12.8 %. These results suggest that EEMF combined with chemometrics and ensemble methods can be used as a promising tool for rapid authenticity identification in commercial edible vegetable oils.