Repositori URV

Identificador: imarina:9280616

Handle: https://hdl.handle.net/20.500.11797/imarina9280616

Autores:
Rovira, GloriaMiaw, Carolina Sheng WheiMartins, Mario Lucio CamposSena, Marcelo Martinsde Souza, Scheilla Vitorino CarvalhoRuisanchez, ItziarCallao, M Pilar

Resumen:
An untargeted strategy was developed to determine cashew nuts adulteration with Brazilian nuts, pecan nuts, macadamia nuts and peanuts. A one-class SIMCA model was developed for the cashew non-adulterated samples by means of two spectroscopic techniques: Near-Infrared (NIR) and Attenuated Total Reflection-Fourier Transform Infrared (ATR-FTIR). Receiver operating characteristic (ROC) curves have been proved to be useful to optimize class limits, both for the NIR and ATR-FTIR models, allowing to balance the values of the performance parameters. An increase in the sensitivity of the training and test set has been obtained from 79% with NIR and 85% with ATR-FTIR to 93% in both cases. As a result, the specificity has slightly decreased from 100% with NIR and a range of 90–98% with ATR-FTIR to a range of 82–98% and 84–96%, respectively. The implementation of high-level data fusion to the classification results obtained from NIR and ATR-FTIR, considering the limit value optimized by ROC curves, allowed the improvement of the performance parameters of the untargeted strategy. Obtaining sensitivity values for the training and test set of 100% and 93%, respectively. Specificity values of 100% were obtained for the detection of Brazilian nuts, macadamia nuts and peanuts, while for pecans it was 98%.