Number of sources author M.L. scored 4 or 5 in the evaluation subset when ordered by the predicted score from several algorithms: random ordering (Random), ordering by number of components in an island from PYBDSF (Complexity), ordering by ASTRONOMALY using ellipse-based features with an anomaly detection algorithm and active learning (Ellipses), and two cases of ordering using the BYOL features combined with PROTEGE for active learning (BYOL). The numbers in brackets for BYOL refer to the number of sources used to train the algorithm. The 3500 source subset is excluded due to the curve’s similarity to the full dataset curve. The envelope around the lines indicates the standard deviation from 10 repetitions. As it results in the steepest curve, BYOL combined with PROTEGE outperforms all other methods when sufficient training data is used.