These results indicated that accurate classification can be achieved from relatively small regions of the spectra, by means of imposing penalties in the models to reduce the number of explicit variables. The α parameter controls the
mixing between Ridge and LASSO regression. Ridge regression (α¼0) imposes a L2-penalty to the model inducing coefficient shrinkage, while LASSO regression (α¼1) imposes a L1-penalty which expects many predictors to be close to zero and a small subset to be nonzero, providing automatic variable selection [22]. LASSO presents some limitations if there is a group of variables among which the pairwise correlations are very high, because it