[Paper Review] Research and characterisation of blazar candidates among the Fermi/LAT 3FGL catalogue using multivariate classifications

Julien Lefaucheur, Santiago Pita|HAL (Le Centre pour la Communication Scientifique Directe)|Mar 6, 2017

Astrophysics and Cosmic Phenomena4 citations

TL;DR

This study uses multivariate classifiers—boosted decision trees and multilayer perceptron neural networks—on Fermi/LAT 3FGL catalogue data to identify 595 new blazar candidates among unassociated sources and classify 509 BL Lacs and 295 FSRQs from candidates and BCUs. The method achieves a low expected false association rate (~9 out of 425) for high-latitude candidates, significantly enhancing the sample of γ-ray blazars for population studies and multiwavelength follow-up.

ABSTRACT

In the recently published 3FGL catalogue, the Fermi/LAT collaboration reports the detection of $γ$-ray emission from 3034 sources obtained after four years of observations. The nature of 1010 of those sources is unknown, whereas 2023 have well-identified counterparts in other wavelengths. Most of the associated sources are labelled as blazars (1717/2023), but the BL Lacs or FSRQ nature of 573 of these blazars is still undetermined. The aim of this study was two-fold. First, to significantly increase the number of blazar candidates from a search among the large number of Fermi/LAT 3FGL unassociated sources (case A). Second, to determine the BL Lac or FSRQ nature of the blazar candidates, including those determined as such in this work and the blazar candidates of uncertain type (BCU) that are already present in the 3FGL catalogue (case B). For this purpose, multivariate classifiers were trained using samples of labelled sources with no caution flag from the 3FGL catalogue and carefully chosen discriminant parameters. The decisions of the classifiers were combined in order to obtain a high level of source identification along with well controlled numbers of expected false associations. Specifically for case A, dedicated classifications were generated for high and low galactic latitude sources, in addition, the application of classifiers to samples of sources with caution flag was considered separately, and specific performance metrics were estimated. We obtained a sample of 595 blazar candidates among the unassociated sources of the 3FGL catalogue. We also obtained a sample of 509 BL Lacs and 295 FSRQs from the blazar candidates cited above and the BCUs of the 3FGL catalogue. The number of expected false associations is given for different samples of candidates. It is, in particular, notably low ($\sim$9/425) for the sample of high-latitude blazar candidates from case A.

Motivation & Objective

• To significantly increase the number of blazar candidates among the 1010 unassociated Fermi/LAT 3FGL sources with unknown nature.
• To determine the BL Lac or FSRQ classification for blazar candidates, including those of uncertain type (BCUs) in the 3FGL catalogue.
• To reduce false associations through robust multivariate classification techniques with controlled error rates.
• To improve the statistical characterization of the γ-ray blazar population for use in extragalactic background light and blazar sequence studies.

Proposed method

• Trained boosted decision trees and multilayer perceptron neural networks on 3FGL sources with no caution flags, using carefully selected discriminant parameters.
• Applied classifiers separately to high-latitude (|b| > 10°) and low-latitude (|b| ≤ 10°) unassociated sources to account for Galactic plane crowding.
• Combined classifier decisions to maximize identification confidence while minimizing false associations through ensemble voting and performance thresholding.
• Evaluated classifier performance on caution-flagged sources separately, estimating expected false positive rates for different candidate samples.
• Used TMVA (Toolkit for Multivariate Analysis) for classifier training and STILTS for cross-matching and data manipulation.
• Validated results against spectroscopic observations and prior studies to assess consistency and reliability.

Experimental results

Research questions

• RQ1 How many new blazar candidates can be identified among the unassociated Fermi/LAT 3FGL sources using multivariate classification?
• RQ2 What is the expected rate of false associations in the identified blazar candidate samples, particularly at high galactic latitudes?
• RQ3 How well do the multivariate classifiers distinguish between BL Lac and FSRQ types in both newly identified candidates and BCUs?
• RQ4 To what extent do the results agree with independent spectroscopic follow-ups and prior classification studies?
• RQ5 Can spectral and variability parameters improve the separation of BL Lacs from FSRQs compared to variability-only classifiers?

Key findings

• The study identifies 595 new blazar candidates among the unassociated Fermi/LAT 3FGL sources, with a low expected false association rate of approximately 9 out of 425 for the high-latitude sample.
• A total of 509 BL Lacs and 295 FSRQs are identified from the combination of new candidates and BCUs, significantly expanding the known γ-ray blazar population.
• The classifier performance for BL Lacs yields a true positive rate of ~84% and a false positive rate of ~5%, outperforming previous variability-only methods.
• For FSRQs, the method achieves ~69% true positive rate and ~12% false positive rate, with better agreement on spectral parameters than on variability alone.
• Among 31 sources from a spectroscopic campaign, 25 were confirmed as BL Lacs or FSRQs by the classifier, with only two misclassified, indicating strong consistency with observational validation.
• The overlap region in variability space between BL Lacs and FSRQs is better resolved when spectral parameters are included, demonstrating their importance in classification.