An appropriate seafood origin identification is essential for labelling regulation but also economic and ecological issues. Near infrared (NIRS) reflectance spectroscopy was employed to assess the origins of cuttlefish caught from five fishing FAO areas (Adriatic Sea, northeastern and eastern central Atlantic Oceans, and eastern Indian and western central Pacific Oceans). A total of 727 cuttlefishes of the family Sepiidae (Sepia officinalis and Sepiella inermis) were collected with a portable spectrophotometer (902-1680 nm) in a wholesale fish plant. NIR spectra were treated with standard normal variate, detrending, smoothing, and second derivative before performing chemometric approaches. The random forest feature selection procedure was executed to select the most significative wavelengths. The geographical origin classification models were constructed on the most informative bands, applying support vector machine (SVM) and K nearest neighbors algorithms (KNN). The SVM showed the best performance of geographical classification through the hold-out validation according to the overall accuracy (0.92), balanced accuracy (from 0.83 to 1.00), sensitivity (from 0.67 to 1.00), and specificity (from 0.88 to 1.00). Thus, being one of the first studies on cuttlefish traceability using NIRS, the results suggest that this represents a rapid, green, and non-destructive method to support on-site, practical inspection to authenticate geographical origin and to contrast fraudulent activities of cuttlefish mislabeled as local.
Keywords: NIRS; authenticity; cephalopods; machine learning; mislabeling; traceability.