Gradient boosting decision trees classification of blazars of uncertain type in the fourth Fermi-LAT catalogue

ABSTRACT The deepest all-sky survey available in the γ-ray band – last release of Fermi-LAT catalogue (4FGL-DR3) based on data accumulated 12 years contains more than 6600 sources. largest population among sources is blazar subclass 3743, 60.1 per cent which are classified as BL Lacertae objects (BL Lacs) or Flat Spectrum Radio Quasars (FSRQs), while rest listed candidates uncertain type (BCU) their firm optical classification lacking. goal this study to classify BCUs using different machine learning algorithms, trained spectral and temporal properties already Lacs FSRQs. Artificial Neural Networks, XGBoost, LightGBM algorithms employed construct predictive models for BCU classification. Using 18 input parameters 2219 FSRQs, we train (80 sample) test (20 cent) these find that model, state-of-the-art algorithm gradient boosting decision trees, provides highest performance. Based our best 825 Lac 405 FSRQ candidates, however, 190 remain without a clear prediction, but percentage 4FGL reduced 5.1 cent. photon index, synchrotron peak frequency, high-energy frequency large sample used investigate relationship between FSRQs (LBLs, IBLs, HBLs).