f₁ score (left), purity (middle), and completeness (right) versus limiting magnitude for the test sample. The thick solid lines show the performance of the combined algorithm, to be compared with the performance of the individual classifiers (thin lines). The vertical dotted lines show the magnitude limit of the miniJPAS survey (Bonoli et al. 2021). The top panel shows the performance for high-z quasars (with z ≥ 2.1) and the bottom panel the performance for low-z quasars (with z < 2.1). We see that the combined algorithm outperforms the rest.