Search for high-confidence blazar candidates and their MWL counterparts in the Fermi-LAT catalog using machine learning

Abstract

A large fraction of the gamma-ray sources presented in the Third Fermi-LAT source catalog (3FGL) is affiliated with counterparts and source types, but 1010 sources remain unassociated and 573 sources are associated with active galaxies of uncertain type. The purpose of this study is to assign blazar classes to these unassociated and uncertain sources, and to link counterparts to the unassociated. A machine learning algorithm is used for the classification, based on properties extracted from the 3FGL, an infrared and an X-ray catalog. To estimate the reliability of the classification, performance measures are considered through validation techniques. The classification yielded purity values around 90% with efficiency values of roughly 50%. The prediction of high-confidence blazar candidates has been conducted successfully, and the possibility to link counterparts in the same procedure has been proven. These findings confirm the relevance of this novel multiwavelength approach.