Near infrared
spectroscopic fingerprints of Corsican honey samples were analysed by a range of chemometric tools to confirm their claimed provenance. Authentic, unfiltered honeys (
n = 373; 219 Corsican and 154 non-Corsican) were collected over 2 production seasons, the goal being to create a specific spectral fingerprint for Corsican honey. Following preliminary data examination by principal component analysis, the multivariate method studied for provenance confirmation was partial least squares regression; various spectral pre-treatments were investigated. Best PLS discriminant models developed using full cross-validation, a variable selection algorithm and a 2nd derivative data pre-treatment, gave correct classification results of 90.0% and 90.3% for the Corsican and non-Corsican honey samples respectively. Using separate calibration and validation samples from this same honey collection, highest correct classification values of 90.4% and 86.3% for Corsican and non-Corsican honey samples respectively were obtained again using a variable selection procedure.
