Performance of the RF model, calculated on the blind test set, applied to different subsets of features. RF_ΦLAB uses the 32 features selected by ΦLAB, RND uses a set of 32 features randomly extracted from the original parameter space (best value over the 10 extractions), while B+W uses the best 10 features plus the 22 excluded by ΦLAB. In all such four parameter spaces, both spectroscopic and photometric redshifts were excluded.