Combination of complementary data mining methods for geographical characterization of extra virgin olive oils based on mineral composition

Highlights

• ICP-MS/OES was applied for multi-element fingerprinting of olive oil samples.

• Complementary data mining methods were used to assess geographical traceability.

• Spanish EVOOs show characteristic mineral profiles according to their origin.

Abstract

This work explores the potential of multi-element fingerprinting in combination with advanced data mining strategies to assess the geographical origin of extra virgin olive oil samples. For this purpose, the concentrations of 55 elements were determined in 125 oil samples from multiple Spanish geographic areas. Several unsupervised and supervised multivariate statistical techniques were used to build classification models and investigate the relationship between mineral composition of olive oils and their provenance. Results showed that Spanish extra virgin olive oils exhibit characteristic element profiles, which can be differentiated on the basis of their origin in accordance with three geographical areas: Atlantic coast (Huelva province), Mediterranean coast and inland regions. Furthermore, statistical modelling yielded high sensitivity and specificity, principally when random forest and support vector machines were employed, thus demonstrating the utility of these techniques in food traceability and authenticity research.

Introduction

Olive oil is the main source of fat in the Mediterranean diet, which has historically been associated with beneficial effects on health. In this sense, the epidemiology suggests that olive oil might have a role in the prevention of coronary diseases and several cancer types, because of its high levels of monounsaturated fatty acids and polyphenolic compounds (Aguilera et al., 2016, Martínez-González and Sanchez-Villegas, 2004). Thereby, the consumption of olive oil has increased in last years, due to its organoleptic and nutritional properties as well as healthy reputation.

Quality, safety and sensorial attributes, such as colour, flavour and taste, are the most important characteristics that determine the commercial value of a food such as olive oil. These features are determined by the chemical composition that, in turn, is affected by pre-harvest and post-harvest factors. Among these pre-harvest factors, olive variety is the main source of variation in composition and sensorial attributes (Aparicio & Harwood, 2013). In addition, quality is also influenced by crop conditions including environmental (climate, soil composition) and agronomic (irrigation, fertilization) factors, sampling time and degree of ripeness (Agiomyrgianaki et al., 2012, Longobardi et al., 2012a, Mihailova et al., 2015, Romero et al., 2016). On the other hand, manufacturing methods used for olive oil extraction and storage conditions are the most important post-harvest factors impacting quality (Ben-Hassine et al., 2013). All these variables are closely associated with the geographical origin and provoke significant differences in organoleptic characteristics, nutritional composition and nutraceutical value of olive oils. Therefore, the traceability and authenticity of olive oil is an important objective in guaranteeing the quality demanded by consumers, producers and regulatory bodies for this valuable food product.

Numerous analytical methodologies have been proposed for the differentiation and classification of olive oils according to the geographical origin, which are usually focused on the determination of different organic compounds (Luykx & van Ruth, 2008). For instance, the profile of fatty acids and triglycerides has previously been used to compare Tunisian, Maghrebian and French virgin olive oils (Laroussi-Mezghani et al., 2015), to characterize Apulian virgin olive oils (Longobardi, Ventrella, Casiello, Sacco, Catucci, et al., 2012) or to distinguish different Turkish oils (Arslan, Karabekir, & Schreiner, 2013). Furthermore, the analysis of minor metabolites such as free sterols, aliphatic and terpenic alcohols, phenolic compounds, tocopherols, hydrocarbons and pigments has also successfully been applied for the classification of olive oils according to their geographical origin (Alonso-Salces et al., 2010, Arslan et al., 2013, Longobardi et al., 2012b). Alternatively, authentication can also be carried out by evaluating the mineral composition of olive oils, which is very informative since it is affected by multiple factors. First, the multi-element profile of plant-derived products primarily depends on the biological demand of the plant as well as on the bioavailability and mobility of mineral compounds from soil. However, some elements can also modify their concentrations in response to agronomic practices (e.g. use of fertilizers and pesticides), climatic factors and manufacturing processes. Thus, mineral elements can be considered as good markers for tracing the geographical origin of virgin olive oils. Furthermore, it should be noted that metal content can be determined by using simple and rapid analytical techniques, unlike tedious time-consuming chromatographic methods needed to analyze organic compounds. Nowadays, inductively coupled plasma mass spectrometry (ICP-MS) and inductively coupled plasma optical emission spectrometry (ICP-OES) are the most commonly used analytical platforms for obtaining elemental fingerprints. In this context, these techniques have successfully been applied to differentiate olive oils from eight European sites (Camin, Larcher, Nicolini, et al., 2010), four Western Greek islands (Karabagias et al., 2013), and four municipalities from Huelva province, in Southern Spain (Beltrán, Sánchez-Astudillo, Aparicio, & García-González, 2015).

High throughput analytical techniques, such as ICP-MS/OES, generate large data sets that require the use of advanced chemometric tools in order to extract the maximum amount of useful information. Several supervised pattern recognition procedures, such as linear discriminant analysis, soft independent model class analogy or partial least squares discriminant analysis, have frequently been used in food analysis to solve authentication problems (Berrueta et al., 2007, Roberts and Cozzolino, 2016). Complementarily, new machine learning algorithms such as random forest, artificial neural networks and support vector machines have demonstrated excellent performance during the last decade for the analysis of complex datasets in many research areas, including food science (Batista et al., 2012, Palacios-Morillo et al., 2016).

The aim of this study was to evaluate the performance of several machine-learning algorithms to discriminate extra virgin olive oils (EVOOs) from different origins by investigating the multi-elemental profile. For this purpose, the present work considers 125 samples from multiple locations across the Spanish geography, which were fingerprinted by using ICP-MS and ICP-OES. Then, three complementary supervised pattern recognition techniques, including partial least squares discriminant analysis (PLS-DA), support vector machine (SVM) and random forest (RF), were applied in order to build statistical models with the aim to discriminate between olive oil groups.