Chemometric analysis of chemo-optical data for the assessment of olive oil blended with hazelnut oil

Abstract

The main objective of this study was to determine different hazelnut oil concentrations in extra virgin olive oil (EV00) belonging to different geographical regions inside Turkey using the combination of a SAW sensor based electronic nose (e-nose) and a machine vision system (MVS). We leveraged the oil characterisation given by the two easy-to-use and complementary experimental techniques through the adoption of conventional PCA for data exploration and random forests (RF) for supervised learning. The e-nose/MVS combination allows significantly better results both in adulteration detection independently of EVOO's geographical provenance and in EVO0 geographical provenance determination, independently of the adulteration level, with respect to the single characterisation method. RF analysis also produces feature ranking, permitting to shed light on which oils' characteristics influence the learning result. We found that EV00 geographical provenance discrimination is mainly due to yellowness and guaiacol content, while (E)-2-hexenal chiefly determines the prediction of the hazelnut level.