Precision (dashed black line) and recall as a function of the classification probability threshold τ for the reverse selection method (Sect. 5.5) over five runs (with randomly chosen training-test splits). The point and error bars represent respectively the mean and the standard deviation over the five runs. The recall lines are calculated considering the whole sample of high-z QSOs with z > 2.5 (blue line) and its subsample of QSOs with z > 3 (red line). The recall curves peak, and have the smaller scatter, at a value of τ = 0.9, hence this is the value adopted as threshold for the analysis in Sect. 5.5.