first two Principal Components (PCs), obtained using the NIR data after MSC and first derivative (Fig. 2) and the UV–visible data after second derivative of the profiles (Fig. 3) show a good distinction between oils from Chianti Classico and oils from Riviera dei Fiori.
The LDA and UNEQ-QDA results are reported in Table 3:in column 1, the 14 matrices studied, with different pre-treatments and transforms, are listed and in columns 2, the number of vari-ables selected by STEP-LDA for each matrix is reported. The LDA prediction abilities (column 3) and the specificities of the QDA-UNEQ models (column 4) are mean values, i.e. they were calcu-lated as average between the values obtained for the two classes Chianti Classico and Liguria Riviera dei Fiori.
The LDA prediction abilities were always high: 100% for 11 over 14 matrices studied.
The specificity is the percentage of the objects of other cate-gories rejected by the model of the studied class; it is important to underline that these specificities refer to models “forced” to have 100% sensitivity, i.e. to models that accept all objects of their own class.