Learning algorithms at the service of WISE survey

TL;DR

This paper presents a machine learning-based classification of nearly 170 million sources in the WISE and SuperCOSMOS datasets, achieving over 96% purity and completeness in identifying galaxies, quasars, and stars for astrophysical analyses.

Contribution

The study introduces a Support Vector Machines classifier trained on SDSS data to classify sources in the WISE and SuperCOSMOS datasets, enabling high-purity, large-scale astronomical source catalogs.

Findings

Achieved over 96% purity and completeness in source classification.

Created a reliable all-sky catalog of ~170 million sources.

Enabled potential for accurate photometric redshift estimation.

Abstract

We have undertaken a dedicated program of automatic source classification in the WISE database merged with SuperCOSMOS scans, comprehensively identifying galaxies, quasars and stars on most of the unconfused sky. We use the Support Vector Machines classifier for that purpose, trained on SDSS spectroscopic data. The classification has been applied to a photometric dataset based on all-sky WISE 3.4 and 4.6 $\mu$m information cross-matched with SuperCOSMOS B and R bands, which provides a reliable sample of $\sim170$ million sources, including galaxies at $z_{\rm med}\sim0.2$, as well as quasars and stars. The results of our classification method show very high purity and completeness (more than 96\%) of the separated sources, and the resultant catalogs can be used for sophisticated analyses, such as generating all-sky photometric redshifts.