SIMBAD references

In this work, we assess the combined use of Gaia photometry and astrometry with infrared data from CatWISE in improving the identification of extragalactic sources compared to the classification obtained using Gaia data. Here we perform a comprehensive study in which we assess different input feature configurations and prior functions to identify extragalactic sources in Gaia, with the aim of presenting a classification methodology that integrates prior knowledge stemming from realistic class distributions in the Universe. In our work, we compare different classifiers, namely Gaussian mixture models (GMMs) and the boosted decision trees, XGBoost and CatBoost, in a supervised approach, and classify sources into three classes, namely star, quasar, and galaxy, with the target quasar and galaxy class labels obtained from the Sloan Digital Sky Survey Data release 16 (SDSS16) and the star label from Gaia EDR3. In our approach, we adjust the posterior probabilities to reflect the intrinsic distribution of extragalactic sources in the Universe via a prior function. In particular, we introduce two priors, a global prior reflecting the overall rarity of quasars and galaxies, and a mixed prior that incorporates in addition the distribution of the extragalactic sources as a function of Galactic latitude and magnitude. Our best classification performances, in terms of completeness and purity of the extragalactic classes, namely the galaxy and quasar classes, are achieved using the mixed prior for sources at high latitudes and in the magnitude range G = 18.5-19.5. We apply our identified best-performing classifier to three application datasets from Gaia Data Release 3 (GDR3), and find that the global prior is more conservative in what it considers to be a quasar or a galaxy compared to the mixed prior. In particular, when applied to the quasar and galaxy candidate tables from GDR3, the classifier using a global prior achieves purities of 55% for quasars and 93% for galaxies, and purities of 59% and 91%, respectively, using the mixed prior. When compared to the performances obtained on the GDR3 pure quasar and galaxy candidate samples, we reach a higher level of purity, 97% for quasars and 99.9% for galaxies using the global prior, and purities of 96% and 99%, respectively, using the mixed prior. When refining the GDR3 candidate tables via a cross-match with SDSS DR16 confirmed quasars and galaxies, the classifier reaches purities of 99.8% for quasars and 99.9% for galaxies using a global prior, and 99.9% and 99.9% using the mixed prior. We conclude our work by discussing the importance of applying adjusted priors that portray realistic class distributions in the Universe and the effect of introducing infrared data as ancillary inputs in the identification of extragalactic sources.