3.1.12. Multivariate Statistical Analysis

A differentiation model built by SLDA classified correctly all the monovarietal EVOOs according to variety (Figure 1) and extracted 30 variables (Table S2). A 100% correct classification was obtained after including 22 variables. Phenol was included in the model as the first and classified correctly 41.76% of all the investigated EVOO samples. After subsequently including nonanoic acid, α-muurolene, 3,7-decadiene I, and estragole (five compounds in total), the total percentage of the correctly classified EVOOs increased to 94.51%.

α

**Figure 1.** Separation of monovarietal extra virgin olive oils produced from Buža, Istarska bjelica, Rosinjola, Oblica, Lastovka, and Leccino varieties in Croatia according to variety in three-dimensional space defined by the first three discriminant functions (roots) on the basis of volatile aroma compound composition.

Unsaturated hydrocarbons from the LOX pathway, accompanied by the most important LOX volatile (*Z*)-3-hexenal, were characterized by the highest positive VIP scores obtained by PLSDA and were confirmed to be typical for Buža EVOO (Figure 2a). Middle-chain fatty acids and hexenol acetates were the volatiles with the highest VIP scores in I. bjelica EVOO (Figure 2b), while the markers of Rosinjola EVOO were mostly benzenoids (Figure 2c). The volatile compound with by far the highest VIP score for the discrimination of Oblica was methyl benzene (Figure 2d), while Lastovka EVOO typicity was mostly owed to its abundance in terpenes and deficiency in LOX volatiles (Figure 2e). High concentrations of benzyl nitrile, (*E*)-2-hexenal, and (*E*,*E*)-2,4-heptadienal, as well as low concentrations of sesquiterpenes, were confirmed to be the most prominent typical characteristics of the Leccino EVOO volatile profile (Figure 2f).

Buža, **Figure 2.** Variable Importance in Projection (VIP) scores of the variables (volatile compounds) most useful for the differentiation of each monovarietal extra virgin olive oil (EVOO), specifically: (**a**) Buža, (**b**) Istarska bjelica, (**c**) Rosinjola, (**d**) Oblica, (**e**) Lastovka, and (**f**) Leccino from the other five EVOOs produced in Croatia. Variables were extracted by partial least squares discriminant analysis applied on mean-centered data of six separate datasets each including two groups (a single vs. other five monovarietal EVOO).
