Distribution of incorrect classifications for the I_E – Y_E, J_E – H_E method of B20 ((left column) and the u – I_E, I_E – J_E method of B20 (centre column) To allow a direct comparison between the u – I_E, I_E – J_E and our machine-learning selection method, we show results from our pipeline under conditions equivalent to those used for the B20 u – I_E, I_E – J_E method: Galaxy photometric redshifts are included as a feature to be trained on; only galaxies detected in u, I_E, and J_E are used; only galaxies in the redshift ranges 0 < z < 1 and 0 < z < 1.5 are used for the Wide and Deep catalogues, respectively. In each panel, we include the values of the F1-score, precision, recall, and @ metrics.