1H-NMR Fingerprinting and Pattern Recognition Stepwise Strategy for Quality and Authenticity Control of Olive Oil

Abstract

The high price of olive oil, its distinctive sensory profile and its reputation as a healthy source of dietary fats make olive oil a target for fraud. The most common types of olive oil fraud are illegal blending with other vegetable oils (VOs) or low-quality olive oils, deliberate mislabelling of less expensive classes of olive oils, other VOs or their blends with olive oils, and mislabelling of geographical origin or Protected Designation of Origin (PDO) declaration. Olive oil adulteration, being one of the biggest financial frauds in the agricultural sector, evidenced the need to update and harmonize analytical methods for quality and authenticity control of olive oil. A novel stepwise strategy based on the 1H-NMR fingerprint of edible oils and multivariate data analysis has been developed in order to assure the authenticity and traceability of olive oils and their declared blends with VOs. This approach provides an analytical tool to detect fraud when olive oil is illegally blended with VOs or a ?legal? blend is falsely labelled respect to the botanical nature of the oils mixed and/or the percentage of each oil in the declared mixture. 1H-NMR spectral data of olive and virgin olive oils and their mixtures with the VOs most commonly used to make blends, i.e., sunflower, high oleic sunflower, corn, virgin and refined avocado, virgin and refined hazelnut, soybean oil, refined palm olein and desterolised high oleic sunflower oils, was analysed by pattern recognition techniques to develop multivariate classification and regression models, which are organised in a decision tree to afford a stepwise strategy for the aimed purposes. Partial least squares discriminant analysis (PLS-DA) provided satisfactory, stable and robust binary classification models with recognition and prediction abilities of 90100% of correct hits for most of the models to distinguish the type of olive oil and identify the VO in the blend. Partial least squares regression (PLS-R) afforded regression models with excellent precisions and acceptable accuracies to determine the percentage of VO in the mixture. The effectiveness of the proposed strategy was tested with blind samples, the results of which were satisfactory and confirmed its potential to support regulations and control bodies.