Raw spectral files were imported into V-PARVUS package [24] for data analysis. Before classification and class-modelling, several pre-treatments and transforms were performed on the three matrices (NOSE, UV-Visible and NIR) as row profiles (below in-dicated as “Profiles”), logarithmic extended multiplicative scat-tering correction (MSC), first and second derivatives. Row profiles were obtained for each sample by dividing the original values of absorbance or transmittance (in the case of spectroscopic techni-ques) or m/z (for the artificial nose) by the sum of the row values; the result is a “percentage signal”.
First and second derivatives of the signals were calculated by means of third-order smoothing polynomials through eleven points.
As a consequence of the several pre-treatments applied, alone or in combination, onto the 3 raw data matrices, a total number of 14 data matrices (3 raw data matricesþ11 pre-treated data ma-trices) were obtained.