Progress of machine-learning-based classification of initial candidates, as is described in Section 2.1. Upper panel: number of visually vetted real moving objects and artifacts at each iteration, from the pool of 1 271 921 initial candidates. Lower panel: threshold probability corresponding to 100 previously unclassified candidates with the highest score produced by the random forest binary classifier trained at each iteration. These candidates are then visually checked and used to improve the classifier on the next iteration. The procedure stops when the threshold probability falls to zero (the classifier stops producing new candidates for visual inspection).