USING SUPERVISED MACHINE LEARNING MODELS TO FIND ACTIVE GALACTIC NUCLEI IN MULTIWAVELENGTH DATASETS

Abstract

Active galactic nuclei (AGN) are unique astronomical sources that emit intense radiation over the entire electromagnetic (EM) spectrum from radio to gamma ray frequencies. They are estimated to be vastly under classified with only 0.1% of them predicted to have been discovered (Padovani). A major contributing factor to this is that in the past high-quality data was not available for sources across multiple wavelengths. It is often not possible to classify a body as an AGN looking at just one section of the EM spectrum due to the variance in emissions from different sources. This implies looking at just one wavelength will lead to under-coverage in all but one type of AGN. With the emergence of new telescopes and technologies, this paper has managed to collate multi-wavelength data on galactic and extragalactic sources from the XMM-Newton and Gaia, and eROSITA telescopes. Applying various supervised learning models, such as random forests and histogram-based boosting on these sources, has allowed us to classify the AGN in these surveys with a ~97% accuracy using only emission data i.e. without including redshifts. These findings indicate that applying similar models to data collected from these telescopes should help overcome the current under-coverage in AGN.