The XXL survey: LV. Galaxy cluster classification from the XXL X-ray source catalogue using a Gaussian process binary classifier trained on imperfectly labelled data

ABSTRACT We present a Gaussian process binary classifier designed to incorporate label uncertainty in its training data, with the aim of selecting galaxy cluster candidates based on their observed X-ray properties. The classifier was trained using sources from the North and South fields of the XXL survey, with label uncertainty derived from the existing XXL galaxy cluster selection criteria. To prevent the classifier from simply replicating the existing XXL selection, we excluded the two X-ray properties originally used by XXL to identify clusters. Applying the classifier to the XXL North catalogue yielded a new sample of 623 candidate sources, recovering 225 of the 248 clusters previously identified by the standard XXL method. We validated the classifier using two independent optically selected cluster samples. Visual inspection of 530 candidates confirmed 271 cluster candidates, including 95 not previously selected by the XXL process. Accounting for 93 uninspected sources, the purity of the sample was estimated at 0.47 $\pm$ 0.02. The newly identified candidates often showed different X-ray morphologies compared to those previously selected by XXL, typically lacking a dominant X-ray component following a $\beta$-model surface brightness profile. While classifier results were robust to being trained on the North or South XXL catalogues, subtle and unresolved differences in behaviour were identified, possibly due to differences in the properties of the two fields (e.g. Galactic column and foreground differences, or time-varying instrument calibration or background characteristics). Overall, we find that the classifier is complementary to the standard XXL processing.