Classifying X-ray Sources Using Machine Learning

Astronomy is entering an unprecedented era with next generation observatories like eROSITA, LSST, SKA, and CTA. The growing volume of astronomical data demand analysis of large amounts of multiwavelength data and the timely classification of numerous sources. Machine Learning algorithms provide an efficient way of identifying the astrophysical nature of hundreds of thousands of unclassified sources in large catalogs. We present the automated multi-wavelength machine-learning classification pipeline (MUWCLASS). The pipeline uses supervised learning with the random forest algorithm. The training datasets are constructed from reliably classified published sources with counterparts in two modern X-ray surveys (Chandra Source Catalog release 2.0 and 4XMM-DR10). The positions of these sources are then cross-correlated with multiwavelength catalogs (e.g., Gaia eDR3, 2MASS, WISE). Major recent upgrades to our ML pipeline include taking into account the measurement uncertainties, applying the variable extinction/absorption corrections, the inclusion of temporal information, and accounting for confusion. An evaluation of the performance of MUWCLASS will be demonstrated. We envision a wide range of potential applications of the MUWCLASS pipeline by the high energy astrophysics community, including aiding in classification of high-energy gamma-ray sources observed in X-rays, classifying the numerous sources in stellar clusters, finding neutron stars in supernova remnants, performing population studies of large numbers of X-ray sources, and classifying eRosita survey sources.