Search for high confidence AGN candidates and its counterparts in the Fermi-LAT unassociated sample using machine learning

The third Fermi-LAT source catalog (3FGL) is the deepest all-sky survey in gamma-rays and comprises 3033 point sources. While for 2023 sources plausible associations have been found, 1010 remain unassociated. A search for active galactic nuclei (AGN) will help to reduce the number of unassociated sources, and will increase our knowledge of the population of gamma-ray emitting AGN. Several machine learning approaches applied to Fermi data have shown the capability of this method. The extension to multiwavelength data improves these studies, and at the same time offers the possibility to determine the most likely corresponding counterpart. As the 95% confidence region of the localization by the Fermi measurement is in the order of several arcminutes, generally multiple point sources at different wavelengths are located within this region and the association is ambiguous. To figure out the most likely counterpart, the associated sample is used to train machine learning classifiers as e.g. the random forest. Therefore, all possible combinations of the Fermi measurement and the measurements at a second wavelength are considered for a particular source. In this talk, the statistical model to obtain high confidence AGN counterpart candidates is described as well as the validation of the model to estimate the performance.