Robust application of a chemometric model based on the relationships between 10 volatile compounds and sensory attributes to support the panel test in virgin olive oil quality classification in olive oil companies

Abstract

Several approaches have been proposed to support the panel test in virgin olive oil classification, but none of them is currently applied in olive oil companies. Aim of this study was the robust application of a chemometric model in a big olive oil company. The application on 244 samples of the PCA-LDA model developed in 2019, based on volatile profile by HS-SPME-GC-MS, gave unsatisfactory results, pointing out critical issues relating to the training-set, variable selection and validation. Therefore, a new t-test-FwS-LDA model was developed; it was based on a very wide dataset (approx. 1800 samples from 6 different production years) and on an algorithm for a stepwise selection of variables. The crucial role of the production year has been proven and included in the model. Ten volatile molecules were thus selected coming from both the lipoxygenase pathway and several virgin olive oil sensory defects. The new model was two-fold validated with 53 and 273 samples coming from production years belonging and not belonging to the training-set, respectively, with very satisfactory results (>90 % and 80 % correct classification, respectively). Finally, the study indicated that for routinary application of the model, year-by-year updating of training-set and variable selection is required.