Research and characterization of blazar candidates among the Fermi/LAT4FGL catalog using Machine Learning

Several machine-learning algorithms were tested to identify blazar candidates among the 4FGL unassociated sources. Using the Scikit-learn toolkit, for a given set of discriminant parameters, methods based on random forests, multilayer perceptions, and support vector machines could reach comparable performance with a very little tuning. We obtain 510 blazar candidates, by combining the results of three classifiers.