Blazar Flaring Patterns (B-FlaP): Classifying Blazar Candidates of Uncertain type in the third Fermi-LAT catalog by Artificial Neural Networks

TL;DR

This paper presents a machine learning approach using Artificial Neural Networks and ECDF to classify uncertain blazar sources in the Fermi-LAT catalog based solely on gamma-ray data, aiding in the identification of blazar subclasses.

Contribution

The study introduces a novel ANN-based classification method for BCUs using gamma-ray data, validated with radio and optical observations, improving the efficiency of blazar identification.

Findings

Classified 342 BCUs as BL Lacs

Classified 154 BCUs as FSRQs

77 BCUs remain uncertain

Abstract

The Fermi Large Area Telescope (LAT) is currently the most important facility for investigating the GeV $\gamma$-ray sky. With Fermi LAT more than three thousand $\gamma$-ray sources have been discovered so far. 1144 ($\sim40\%$) of the sources are active galaxies of the blazar class, and 573 ($\sim20\%$) are listed as Blazar Candidate of Uncertain type (BCU), or sources without a conclusive classification. We use the Empirical Cumulative Distribution Functions (ECDF) and the Artificial Neural Networks (ANN) for a fast method of screening and classification for BCUs based on data collected at $\gamma$-ray energies only, when rigorous multiwavelength analysis is not available. Based on our method, we classify 342 BCUs as BL Lacs and 154 as FSRQs, while 77 objects remain uncertain. Moreover, radio analysis and direct observations in ground-based optical observatories are used as counterparts to the statistical classifications to validate the method. This approach is of interest because of the increasing number of unclassified sources in Fermi catalogs and because blazars and in particular their subclass High Synchrotron Peak (HSP) objects are the main targets of atmospheric Cherenkov telescopes.