Confusion matrix for the ANN classifier used to select LAEs. The three classes considered are: QSOs selected by a feature other than Lyα (“QSO cont.”), QSOs correctly selected by their Lyα line, in the correct redshift (“QSO LAE”), and low-z (z ≈ 0–2) galaxy contaminants (“Low−z galaxy”). The annotations in the cells are the normalized counts so the vertical columns add up to 1. We show the confusion matrices for the full test set (top, L_Lyα > 10^43.5 erg s⁻¹) and splitting in low (middle, L_Lyα < 10⁴⁴ erg s⁻¹) and high luminosity bins (bottom, L_Lyα > 10⁴⁴ erg s⁻¹). This figure shows that the ANN effectively selects QSO LAEs with an accuracy of 93% for the general sample, while maintaining a low false negative rate (6% and 1% in the case of objects labeled as QSO cont. and low-z galaxies, respectively).