Presentation #101.21 in the session AGN & Quasars — iPoster Session.
Transparent machine learning models are those models whose inner workings are understandable by people. This study uses several machine learning models, Decision Trees, Random Forests, Support Vector Machines, and K Nearest Neighbors to classify a sample of 529 Active Galactic Nuclei (AGN). These AGN are drawn from the MOJAVE (N = 231) survey and from FERMI/LAT (N = 298). Web-scrapping from NASA Extragalactic Database is used to obtain photometric data at 1.4 GHz, 4.85 GHz, 3.4 micrometers (WISE1) and 0.1-100 GeV (Fermi/LAT). Training and test sets are taken at random from the parent sample and cross validation is used to evaluate the effect of random sampling. We systematically change internal parameters, such as the kernel of the Support Vector Machine or the model depth of a Decision Tree to see which, if any gives the best results. The various models are then compared.