22mg kg -1 , 29 were correctly predicted and 1 sample was incorrectly predicted.Only 2 samples were considered as not belonging to any of the classes, and thus the model presented satisfactory results. In the case of PLS-DA, the average accuracy was 77%, and 3 latent variables were selected for each class. The KNN and SIMCA chemometric strategies presented good results for the two classes; however, the KNN model showed better accuracy (98% on average). Table 3 shows the results for KNN, SIMCA and PLS-DA and some figures of merit (%) for the proposed models (accuracy, sensitivity, false alarm rate and specificity). 42As can be noted in Table 3, the accuracy for KNN and SIMCA ranged from 96 to 98%. This parameter is related to the number of correct predictions. The average sensitivity for KNN was 97%. This parameter is related to the samples that belong to a certain class and were predicted to belong to the other class. The sensitivity for class 1 (below) in KNN, for example, was 94%, which means that one sample from class 2 (above) was predicted as class 1. On the other hand, the sensitivity for class 2 was 100%, which means that no sample from class 2 was predicted as class 1.The average sensitive was 97% ((94+100)/2). Specificity is the inverse of sensitivity, and false alarm rate for class 2, for example, is the number of samples from class 2 predicted as belonging to class 1 divided by the total