Miloš Kovacecvic, Graziano Chiaro, Gino Tosti, Sara Cutini.
In the Fermi-LAT Fourth Source Catalogue (3FGL) about 50% of the sources have no clear association with a likely γ-ray emitter. We use machine learning techinque aimed at distinguishing BL Lacs from FSRQs to investigate the source subclass of uncertain (BCU) or unassociated ( UCS) sources characterised by γ-ray properties very similar to those of Active Galactic Nuclei.
This work is a follow up of previous papers : https://arxiv.org/abs/1607.07822 , https://arxiv.org/abs/1705.09832, https://arxiv.org/abs/1808.05881, https://arxiv.org/abs/1602.00385 and will use the 2019 optimization of the original algorithm as described in : Optimizing neural network techniques in classifyingFermi-LAT-ray sources.
The result of this study will suggest a new zoo for 4FGL γ-ray objects, opening up new considerations on the population of the γ-ray sky, and it will facilitating the planning of significant samples for rigorous analyses and multiwavelength observational campaigns.
4FGL BCUs Classification
Class 1FGL 2FGL 3FGL 4FGL
BL Lac 295 (44%) 436 (41%) 660 (38%) 1116 (36%)
FSRQ 278 (42%) 370 (35%) 484 (28%) 686(22%)
BCU 92 (14%) 257 (24%) 573 (34%) 1329(42%)
Total 665 1063 1717 3131
Table 1. Blazar class distribution in Fermi-LAT catalogs.
Classifying BCUs, using a supervised machine learning method based on an artificial neural network, probabilities for each of 1329 uncertain blazars to be a BL Lac or FSRQ are obtained.
Using 90% precision metric, 801 can be classified as BL Lacs and 406 as FSRQs while 122 still remain unclassified.
Here the full list 4FGL BCUs_ANN_table.txt
A spectrometric optical observation campaign will be organized to confirm the data resulting from the neural algorithm.