Determination of the geographical origin of Italian extra virgin olive oil using pyrolysis mass spectrometry and artificial neural networks

Olives were collected from five important regions of Italy, from as many cultivars and locales as possible. For each region a number of samples were produced, representative of the area as a whole. Once collected the olives were washed and processed using standard methods within the ISE olive mill to produce DOC extra virgin olive oils of known region, province and variety (cultivar). These oils were analysed in triplicate by Curie-point pyrolysis mass spectrometry and the spectra collected. Spectra were normalised and sorted according to region. The data-splitting program, Multiplex (A. Jones, D.B. Kell and J. Rowland, Submitted to Analytica Chimica Acta (1996)) was used to sort the spectra into training and test sets split in the ratio 2:1 for Abruzzo:Sardinia and Apulia:Sardinia predictions and a ratio of 1:1 for Lazio:Sicily. Using artificial neural nets with a single output that represented the network's estimation of the geographical provenance as a numeric code, all unknown samples (as triplicates) from an Abruzzo/Sardinia challenge were successfully identified. Samples form a Lazio/Sicily were successfully predicted/separated when the outputs from the network for each triplicates were averaged, and Apulia/Sardinia were predicted with only a single error for each region. This represents the first report in which the precision and discrimination of pyrolysis mass spectrometry has been shown, when combined with artificial neural networks, to allow the discrimination of olive oils by region. (C) 1997 Elsevier Science B.V.