2025PASP..137f4105D

Gamma-ray sources are among the highest energy phenomena in the Universe and are important laboratories for the study of particle acceleration and relativistic astrophysics. The classification of gamma-ray sources in the past has been aided by the incorporation of multiwavelength observations. In this study, we classify different types of active galactic nuclei by applying machine learning (ML) techniques to a crossmatch of newly released gamma-ray and X-ray catalogs for the first time: the Fermi Gamma-ray Large Area Telescope Fourth Source Catalog and the recent, first release of the SRG/eROSITA All-Sky X-ray Survey. We investigate the effects that different features from this dataset have on the classification accuracy and compare the ML model to traditional methods that use a simple cut in parameter space. We find that a variant of the Gradient Boosting ML model leads to an increased classification accuracy of previously manually classified sources (99.2%) compared to that of traditional methods (92.0%). Our models, trained on gamma-ray and X-ray data, even outperform recent ML models trained on gamma-ray data alone, which achieve accuracies no greater than 92.4%. We also find that the addition of X-ray data to gamma ray data increases both the accuracy and the number of sources classified with high confidence. Finally, we apply our model to blazar candidates of unknown type within the dataset and find 221 sources with a >90% probability of being either a BL Lacertae object or flat spectrum radio quasar. We consider all classifications to be candidates since the labels are derived from the Fermi catalog itself, not optical spectroscopic surveys. With the increasing number of gamma-ray and X-ray detections and large numbers of sources yet to be classified, this work highlights the strong potential of ML in the classification of multi-wavelength astronomical datasets.