Rapid authentication and classification of grape seed oil using fluorescence spectroscopy combined with sparse classification and regression methods

Authentication and classification of grape seed oil (GSO) varieties have attracted much attention in food industry, in recent years [1]. In the present work, excitation-emission fluorescence spectroscopy and sparse chemometric methods were used to classify different varieties of GSO taken from different grape genotypes. Moreover, GSO adulteration with sunflower oil (SFO) was successfully assessed using sparse regression methods. Fluorescence spectra were obtained in the wavelength regions of λex= 200-500 nm and λem= 200-800 nm. More than 200 samples from five different GSO genotypes were used to build multivariate models. The sparse version of N-way partial least squares discriminant analysis (sNPLS-DA) [2] was used to develop interpretable classification models. The capabilities of the sNPLS-DA model provide a valuable insight about the important wavelengths in fluorescence spectra to distinguish between GSOs. Furthermore, adulterant levels in GSO samples were quantified using sparse regression techniques [3], including the least absolute shrinkage and selection operator (Lasso), Ridge, and Elastic net, and the results were compared with those obtained using the VIP-PLS method. The overall accuracy for sNPLS-DA was 1.00 and the coefficient of multiple determination (R2) for Lasso model was 0.914, for the test sets. The results in this work revealed that sparse classification and regression models, including sNPLS-DA and Lasso, coupled with excitation-emission fluorescence spectroscopy can be effectively used to assess the quality of oil samples in the food industry