Comparison of artificial neural networks and multiple regression tools applied to near infrared spectroscopy for predicting sensory properties of products from quality labels

Abstract

In products from quality labels a sensory analysis is obligatory although this is a slow and expensive process. This study examines the prediction of the sensory parameters of chorizo dry-cured sausage by using NIRS technology and the application of chemometric methods such as MPLS (Modified Partial Least Square regression) and ANN (Artificial Neural Networks). The results show that by applying ANN it is possible to predict the 20 sensory parameters analyzed with RSQ values of from 0.61 to 0.92; these values are always higher than those obtained by prediction using MPLS. Moreover, the combination of NIRS and RMS-X residual discrimination allowed the correct classification of 94.4% of the samples according to whether or not they belonged to a certain Quality Label.