Risultati e discussione

The two main parameters that will be used to evaluate the goodness of the regression models VISSWIR spectra and chlorate/perchlorate content are: Root Mean Square Error (RMSE) and 𝑅ଶ ​​(Martens et al., 1989; Naes, 2002). The RMSE parameter provides the possibility to evaluate the differences between the estimated response 𝒀̑ and ​​the observed response 𝒀. While, R2 is a parameter used to evaluate the model fitting. These parameters will be referred to the calibration, cross-validation and prediction phases. Meanwhile, to evaluate the importance of the contribution of the descriptive variables to the final model, both the variables significant to the description of Y must be taken into consideration (and the variables relevant to the description of X, summarized by parameters such as: regression vector, selectivity ratio and VIP scores. The classification performances will instead be evaluated on the basis of the parameters: Sensitivity, Specificity, Precision, Misclassification Error and Accuracy, referred to the calibration, cross-validation and prediction phase of the test set.