Olive oil samples from three different Greek regions (Crete, Peloponnese and Lesvos) were examined by optical spectroscopy in a wide spectral region from ultraviolet to near infrared using absorption, fluorescence and Raman spectroscopies. With the aid of machine learning methods, such as multivariate partial least squares discriminant analysis, a clear classification of samples originating from the different Greek geographical regions was revealed. Moreover, samples produced in different subareas of Crete and Peloponnese were also well discriminated. Furthermore, mixtures of olive oils from different geographical origins were studied employing partial least squares as a tool to establish a model between the actual and predicted compositions of the mixtures. The results demonstrated that optical spectroscopy combined with multivariate statistical analysis can be used as an emerging innovative alternative to the classical analytical methods for the identification of the origin and authenticity of olive oils.
Keywords: Partial Least Squares Discriminant Analysis; Raman spectroscopy; authenticity; geographical origin; machine learning; olive oil; visible spectroscopy.