The problem of authenticating extra virgin olive oil varieties is particularly important from the standpoint of quality control. After having shown in our previous works the possibility of discriminating oils from a single variety using chemometrics, in this study a combination of two different neural networks architectures was employed for the resolution of simulated binary blends of oils from different cultivars. In particular, a Kohonen self-organizing map was used to select the samples to include in the training, test and validation sets, needed to operate the successive calibration stage, which has been carried out by means of several multilayer feed-forward neural networks. The optimal model resulted in a validation Q(2) in the range 0.91-0.96 (10 data sets), corresponding to an average prediction error of about 5-7.5%, which appeared significantly better than in the case of random or Kennard-Stone selection. (C) 2007 Elsevier B.V. All rights reserved.

Use of different artificial neural networks to resolve binary blends of monocultivar Italian olive oils / Marini, Federico; Magri', Antonio; Bucci, Remo; Magri', Andrea. - In: ANALYTICA CHIMICA ACTA. - ISSN 0003-2670. - STAMPA. - 599:2(2007), pp. 232-240. [10.1016/j.aca.2007.08.006]

Use of different artificial neural networks to resolve binary blends of monocultivar Italian olive oils

MARINI, Federico;MAGRI', Antonio;BUCCI, Remo;MAGRI', Andrea
2007

Abstract

The problem of authenticating extra virgin olive oil varieties is particularly important from the standpoint of quality control. After having shown in our previous works the possibility of discriminating oils from a single variety using chemometrics, in this study a combination of two different neural networks architectures was employed for the resolution of simulated binary blends of oils from different cultivars. In particular, a Kohonen self-organizing map was used to select the samples to include in the training, test and validation sets, needed to operate the successive calibration stage, which has been carried out by means of several multilayer feed-forward neural networks. The optimal model resulted in a validation Q(2) in the range 0.91-0.96 (10 data sets), corresponding to an average prediction error of about 5-7.5%, which appeared significantly better than in the case of random or Kennard-Stone selection. (C) 2007 Elsevier B.V. All rights reserved.
File allegati a questo prodotto
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11573/232542
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? 2
  • Scopus 57
  • ???jsp.display-item.citation.isi??? 53
social impact