*3.2. Feature Importance*

–

–

Observing the top 40 MANOVA rotation test's most important variables (e.g., spectral lengths), the most informative ones ranged within the following frequencies: 308–373 nm, 594–612 nm, and 617–641 nm. The average VIS-NIR spectral data of foreign and Italian EVOOs are reported in Figure 5 together with the higher importance spectral values extracted with the aforementioned MANOVA rotation test. 's – – – – – –

's

**Figure 5.** Mean VIS-NIR spectral data: Italian (red line), foreign (blue line). Higher importance spectral values extracted with the MANOVA rotation test are evidenced with green rectangles.

– – – – – Consequently, the 40 most important features extracted through the ANN procedure (e.g., spectral lengths), ranged within the following frequencies: 308–378 nm, 415–422 nm, 474–507 nm, 564–570 nm, and 596–605 nm. The average VIS-NIR spectral data of both Italian and foreign samples, together with the higher importance spectral values in terms of ANN feature importance, are reported in Figure 6. – – – – –

**Figure 6.** Mean VIS-NIR spectral data: Italian (red line), foreign (blue line). Higher importance spectral values in terms of artificial neural network (ANN) feature importance are evidenced with green rectangles.

– – – – The two feature importance approaches, MANOVA and ANN, evidenced common ranges of higher importance, which were: 308–373 nm and 594–605 nm. These spectral bands represent portions of the visible spectral range. The color of an oil is, therefore, due to the combination and proportion of its pigments [49]. These molecules do not depend only on the characteristics of the fruits (*Olea europaea* L.), the extraction processes used to produce the oil, and the conservation conditions [50] but, also, on weather and pedo-climate conditions [51]. Therefore, the relationship between the stage of ripeness and pigment content in EVOO could be, indeed, very important for further authentication studies [52].

The molecular structure of chlorophylls and, in particular, the planar structure of the tetrapyrrolic macrocycle coordinated by a magnesium ion, Mg ++, is responsible for the absorption of visible light in the green region. Chlorophyll a gives a greenish-blue coloration, while chlorophyll b determines a yellowish-green color. The sensitivity of chlorophylls to extreme temperature and pH allows the formation of several distinct derivatives such as pheophytins, chlorophyllides, and pheophorbides. During the olive oil extraction process, the release of acids may cause pheophytinization reactions in the chlorophyll fraction, increasing the oils' pheophytin content. The conversion of chlorophylls to Mg2<sup>+</sup> free derivatives, such as pheophytins, where the Mg ++ ion is replaced by two H<sup>+</sup> ions, causes oil color changes over time [53–55]. Pheophytin a is present in greater quantities than pheophytin b. If olive oil is not well preserved, pheophytins can transform further, degrading to pyro pheophytin [55]. These latter can be considered an index of an aging oil. In addition to chlorophyll derivatives, pigments in extra virgin olive oil include carotenoids, the majority of which are lutein and carotene. Carotenoids are isoprenoid compounds with a hydrocarbon structure with various double bonds, C–C, which are responsible for their interesting properties as antioxidants [56]. Carotenoids can be further divided into carotenes (which contain only carbon and hydrogen atoms) and xanthophylls (which also contain oxygen atoms).

The spectra of olive oils analyzed in this work agree with those reported in the literature [57–61].

The peaks occurring in the range between 308 and 380 nm are mostly due to phenolic components [62]. In detail, we found the peak at around 350 nm, the absorption zone of flavones, present in the EVOO absorbance spectrum useful to distinguish Italian EVOOs from foreign ones. Flavonoids are plant secondary metabolites with different phenolic structures. These compounds are mostly used to generate pigments, which play an important role in the colors of plants producing yellow or red/blue pigmentation. Flavonoids such as apigenin, apigenin-7-*O-*glucoside, luteolin, luteolin-7-*O*-glucoside, luteolin-4-*O*-glucoside, diosmetin, quercetin, and quercetin-3-rutinoside are present in olive oils and contribute to the health benefits of consumers. The antioxidant and cellular damage repairing properties that make them useful for preventing cancer, cardiovascular disease, and degenerative diseases in general have been widely studied [63]. The main factors that contribute to their increase in the oil is the maturity index of the fruits and the degree of grinding and the malaxation conditions of the paste during the extraction processes of the oil [64].

The peaks occurring in the range between 415 and 422 nm are due to the compounds absorbing dark blue colored light, mainly carotenoids, as well as pheophytin a, pheophorbide a, and pyro pheophytin a [59], and are characterized by a yellow color.

The peaks occurring in the range between 474 and 507 nm are due to the compounds absorbing green/yellow colored light, and correspond to carotenoids, such as astaxanthin and canthaxanthin. In any case, the major carotenoids in olive oil are β-carotene and lutein, both of them providing several health benefits. Lutein exhibits antioxidant and anti-inflammatory activity protecting against DNA damage [65].

Moreover, the peaks occurring in the range between 564 and 570 nm and between 594 and 605 nm are due to the compounds absorbing orange colored light, characterized by purple/violet and green/blue colors, respectively, and corresponding to chlorophylls and anthocyanins.

#### **4. Conclusions**

Spectroscopic techniques paired to chemometric analyses are widely used to authenticate and differentiate edible oils. Most spectroscopic methods tend to focus on the major compounds of the saponifiable fraction of an oil, and only a few have been concentrating on the contents of minor compounds, such as pigments and antioxidants. The European community has not yet accepted many of the scientific community's indications concerning minor compounds, which, by law, are not taken into consideration for the definition of EVOOs' authenticity. However, many of the minor compounds are present in significant amounts only in EVOOs, and their quantification could greatly help the oil industry. Although further analysis will be needed to expand the case studies on olive oils, this work provides a clear indication of how pigment and antioxidant contents are crucial for the authentication and definition of the quality parameters of an EVOO. In detail, we found that the peak at about 360 nm and the broad band around 550 nm present in the EVOO absorbance spectrum can be used to distinguish Italian EVOO from foreign ones. As opposite to expensive and time-consuming chromatographic methods, procedures relying on (open source) spectroscopic instruments are cheap (less than 1000 €) and do not require sample preprocessing. Moreover, being fast, these techniques can be used to assess a huge collection of samples within a reasonable time. The quantitative analysis of pigments can take place directly at production sites and stores, through portable tools that are easy to use, even by non-expert staff. The trained ANN used to classify the samples according to their optical

spectra can be easily implemented on an app for immediate classification. The development of simple and reliable methods that can verify the authenticity and guarantee the quality of agri-food products is crucial. Encouragingly, this type of analysis would be very beneficial for the producers themselves as well as consumers. Indeed, these techniques can score comparable precision with respect to the more expensive and time-consuming traditional ones. Moreover, since their application cost relies entirely on the instrumental budget, and not on reagent or other expensive consumable materials, they can be applied to a high number of samples and thus, in case of supposed fraud, can be used as pre-screening tools leading to time and economic optimization.

**Author Contributions:** Conceptualization, F.P. and C.C.; data curation, C.C., S.V., and L.O.; formal analysis, C.C., S.V., and L.O.; funding acquisition, F.P. and C.C.; investigation, S.V., F.P., L.O., C.B., and C.C.; methodology, S.F., S.V., L.O., and C.C.; project administration, C.C.; software, S.F. and L.O.; supervision, F.P. and C.C.; validation, L.O. and C.C.; visualization, L.O., C.B., and C.C.; writing—original draft, S.V., F.A., C.B., F.P., and C.C.; writing—review and editing, F.P. and S.V. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Italian Ministry of Agriculture (MiPAAF), grant number D.M. n.12479 of project INFOLIVA.

**Acknowledgments:** The authors would like to acknowledge all the farms which sent the EVOO to be analyzed and Sara Yazji for illuminating discussions on the VIS-NIR absorbance spectra.

**Conflicts of Interest:** The authors declare no conflict of interest.
