The probabilistic random forest applied to the selection of quasar candidates in the QUBRICS survey

Abstract

The number of known, bright (i < 18), high-redshift (z > 2.5) QSOs in the Southern Hemisphere is considerably lower than the corresponding number in the Northern Hemisphere due to the lack of multi-wavelength surveys at δ < 0. Recent works, such as the QUBRICS survey, successfully identified new, high-redshift QSOs in the South by means of a machine learning approach applied on a large photometric dataset. Building on the success of QUBRICS, we present a new QSO selection method based on the Probabilistic Random Forest (PRF), an improvement of the classic Random Forest algorithm. The PRF takes into account measurement errors, treating input data as probability distribution functions: this allows us to obtain better accuracy and a robust predictive model. We applied the PRF to the same photometric dataset used in QUBRICS, based on the SkyMapper DR1, Gaia DR2, 2MASS, WISE and GALEX databases. The resulting candidate list includes 626 sources with i < 18. We estimate for our proposed algorithm a completeness of ∼84 per cent and a purity of $\sim 78{{\ \rm per\ cent}}$ on the test datasets. Preliminary spectroscopic campaigns allowed us to observe 41 candidates, of which 29 turned out to be z > 2.5 QSOs. The performances of the PRF, currently comparable to those of the CCA, are expected to improve as the number of high-z QSOs available for the training sample grows: results are however already promising, despite this being one of the first applications of this method to an astrophysical context.

The space density of ultra-luminous QSOs at the end of reionization epoch by the QUBRICS Survey and the AGN contribution to the hydrogen ionizing background

“…The QUBRICS team is currently exploring different selection criteria, e.g. the Probabilistic Random Forest (PRF, Guarneri et al 2021) and the Extreme Gradient Boosting (XGB, Calderone et al in prep. ), which could be tuned in order to have high completeness at z > 4.5.…”

“…The QUBRICS survey (Calderone et al 2019;Boutsia et al 2020Boutsia et al , 2021Guarneri et al 2021) turns out to be particularly efficient and complete in the selection of ultra-bright QSOs at high redshift (z > 2.5). Thanks to the extensive spectroscopic confirmations carried out progressing with this survey, and complementing our database with the results of other groups (Wolf et al 2020;Onken et al 2021), a sample of 14 ultra-bright QSOs with M 1450 ≤ −28.3 at 4.5 < z spec < 5.0 has been assembled in Table 1.…”

“…In the meanwhile, other high-z bright QSOs are going to be discovered by QUBRICS, thanks to the exploitation of new state-of-the-art machine learning techniques, e.g. Probabilistic Random Forest (PRF, Guarneri et al 2021) or Extreme Gradient Boosting (XGB, Calderone et al in prep. ).…”

The Quasar Candidates Catalogs of DESI Legacy Imaging Survey Data Release 9

“…In this work, to create a catalog of QSO candidates for DESI, we adopt a machine learning (ML) technology named Random Forest (RF) and apply it to photometry data from DESI-LIS because its efficiency, flexibility, and accuracy have been intensively proved previously (e.g. Viquar et al 2018;Bai et al 2019;Clarke et al 2020;Guarneri et al 2021), in particular, Bai et al (2019) demonstrates that RF is the most efficient and reliable one among several methods in dealing with quasarstar-galaxy classification. The training and validation sets are built upon the spectra data from SDSS eBOSS (extended Baryon Oscillation Spectroscopic Survey, Dawson et al 2016) DR16 and the photometry data from WISE 2 (Wide-field Infrared Survey Explorer, Wrigh et al 2010) and DESI-LIS, labels are generated based on the database of SIMBAD 3 (the Set of Identifications, Measurements and Bibliography for Astronomical Data, Wenger et al 2000).…”

Near-infrared spectroscopy of extreme BAL QSOs from the QUBRICS bright quasar survey

“…The QUBRICS survey ("QUasars as BRIght beacons for Cosmology in the Southern hemisphere") was started in 2018 to even up a significant lack of identified QSOs in the Southern Hemisphere. The project entails the selection of QSO candidates from public databases, using innovative machine-learning techniques (Calderone et al 2019;Guarneri et al 2021) and their spectral confirmation through direct observation in the optical band (Calderone et al 2019;Boutsia et al 2020). The result is a growing catalogue of some 400 newly discovered bright QSOs, which will significantly enhance the feasibility of a redshift drift measurement with future facilities (the so called Sandage Test; Boutsia et al 2020) and were already used to put stronger constraints on the bright end of the QSO luminosity function (Boutsia et al 2021).…”