12-20 July 2017
Asia/Seoul timezone
Home > Timetable > Session details > Contribution details


BEXCO - Room E(106/107)

[GA169] Towards Refined Population Studies: High-Confidence Blazar Candidates and their MWL Counterparts using Machine Learning


  • Sabrina EINECKE

Primary authors



The Large Area Telescope (LAT) on board the Fermi satellite conducted the deepest all-sky survey in gamma rays so far. Despite outstanding achievements in assigning source types, 1010 sources in the Third Fermi-LAT Source Catalog (3FGL) remain without plausible associations, and 573 sources are associated to active galaxies of uncertain type. Assigning blazar classes to unassociated and uncertain sources, and linking counterparts to the unassociated ones, will refine tremendously our knowledge of the population of gamma-ray emitting objects. The application of machine learning algorithms has become an integral part of exploring astrophysical data. Previous machine learning strategies to assign source types were based solely on properties extracted from gamma-ray observations. The extension to multiwavelength information, especially the relation between properties extracted from different parts of the energy spectrum, provides additional source type-specific characteristics for better classification. At the same time, it offers the possibility to determine the most likely corresponding counterpart. The source localization accuracy of Fermi measurements is in the order of several arcminutes. Typically several hundred possible counterparts are located within this region, making the association ambiguous. To figure out the most likely counterpart, the sample of associated 3FGL sources is used to train machine learning classification algorithms. For any particular 3FGL source, all possible combinations with measurements of one additional energy range are considered, e.g. from the Wide-Field Infrared Survey Explorer (WISE) source catalog, the Sydney University Molongo Sky Survey (SUMSS) radio catalog, or the Swift X-ray Point Source (1SXPS) catalog. By merging the most probable candidates of each of those studies, the power of multiwavelength strategies is exploited and conclusions with even higher confidence concerning blazar counterpart candidates are drawn. In this contribution, the statistical model and its validation to estimate the performance is described. Finally, results of the application of this novel wavelength-dependent approach are presented, and its consequences concerning blazar population studies are discussed.