Machine-learning classifiers were trained with Gaia sources selected from over seven
hundred fifty thousand objects crossmatched with the literature, representing a large
number of variability types as well as non-varying objects. The training set included
about thirty-three thousand sources filtered according to their distribution in
the sky, their number of FoV transits, and their median magnitudes in the band,
as described in more details in Section 7.3.3.
All sources with two or more FoV transits in the band were processed by the
classifiers. Photometric time series in the , , and bands were used after
the pre-processing steps described in Section 7.2.3 and
astrometric quantities (such as parallax and proper motion) were employed without
specific selections. The results of the Statistical Parameter Computation module
(Section 7.2.3) provided additional input information
which was used directly as classification attributes or in the computation thereof.