J/ApJ/887/134 Classification of Fermi blazar cand. from the 4FGL (Kang+, 2019)
Evaluating the classification of Fermi BCUs from the 4FGL catalog using machine
learning.
Kang S.-J., Li E., Ou W., Zhu K., Fan J.-H., Wu Q., Yin Y.
<Astrophys. J., 887, 134 (2019)>
=2019ApJ...887..134K 2019ApJ...887..134K
ADC_Keywords: Active gal. nuclei ; Gamma rays ; BL Lac objects ; QSOs;
Models
Keywords: Blazars
Abstract:
The recently published fourth Fermi Large Area Telescope source
catalog (4FGL) reports 5065 gamma-ray sources in terms of direct
observational gamma-ray properties. Among the sources, the largest
population is the active galactic nuclei (AGNs), which consists of
3137 blazars, 42 radio galaxies, and 28 other AGNs. The blazar sample
comprises 694 flat-spectrum radio quasars (FSRQs), 1131 BL Lac-type
objects (BL Lacs), and 1312 blazar candidates of an unknown type
(BCUs). The classification of blazars is difficult using optical
spectroscopy given the limited knowledge with respect to their
intrinsic properties, and the limited availability of astronomical
observations. To overcome these challenges, machine-learning
algorithms are being investigated as alternative approaches. Using the
4FGL catalog, a sample of 3137 Fermi blazars with 23 parameters is
systematically selected. Three established supervised machine-learning
algorithms (random forests (RFs), support vector machines (SVMs),
artificial neural networks (ANNs)) are employed to general predictive
models to classify the BCUs. We analyze the results for all of the
different combinations of parameters. Interestingly, a previously
reported trend the use of more parameters leading to higher accuracy
is not found. Considering the least number of parameters used,
combinations of eight, 12 or 10 parameters in the SVM, ANN, or RF
generated models achieve the highest accuracy (Accuracy ∼91.8%, or
∼92.9%). Using the combined classification results from the optimal
combinations of parameters, 724 BL Lac type candidates and 332 FSRQ
type candidates are predicted; however, 256 remain without a clear
prediction.
File Summary:
--------------------------------------------------------------------------------
FileName Lrecl Records Explanations
--------------------------------------------------------------------------------
ReadMe 80 . This file
table1.dat 80 34 The results of the two-sample test for 1131 BL Lacs
and 694 FSRQs
table2.dat 70 172 The test accuracy, predict results, and parameters
for the optimal combinations
table3.dat 109 1312 The classification of Fermi blazar candidates of
an unknown type (BCUs)
--------------------------------------------------------------------------------
See also:
J/ApJ/700/597 : FERMI LAT detected blazars (Abdo+, 2009)
J/ApJ/716/30 : SED of Fermi bright blazars (Abdo+, 2010)
J/ApJ/715/429 : First Fermi-LAT AGN catalog (1LAC) (Abdo+, 2010)
J/ApJ/743/171 : The 2LAC catalog (Ackermann+, 2011)
J/ApJ/753/83 : Associations to 1FGL sources (Ackermann+, 2012)
J/MNRAS/424/L64 : AGN/pulsar distinction for 2FGL sources (Mirabal+, 2012)
J/MNRAS/428/220 : Gamma-ray AGN type determination (Hassan+, 2013)
J/ApJ/782/41 : 231 AGN candidates from the 2FGL catalog (Doert+, 2014)
J/ApJ/810/14 : 3rd catalog of LAT-detected AGNs (3LAC) (Ackermann+, 2015)
J/MNRAS/451/2750 : Blazar sequence (Xiong+, 2015)
J/MNRAS/450/3568 : Non-Fermi blazar sample (Xiong+, 2015)
J/other/RAA/16.13 : Sample of Fermi Blazars (Chen+, 2016)
J/MNRAS/462/3180 : 3FGL Blazar of Unknown Type classification (Chiaro+, 2016)
J/ApJS/226/20 : X-ray, opt. & radio SEDs of Fermi blazars (Fan+, 2016)
J/ApJ/820/8 : 3FGL sources statistical classif. (Saz Parkinson+, 2016)
J/ApJS/222/24 : VRI LCs of Mrk 501 from 2010 to 2015 (Xiong+, 2016)
J/A+A/602/A86 : Blazar cand. among Fermi/LAT 3FGL cat. (Lefaucheur+, 2017)
J/MNRAS/470/1291 : Classifying 3FGL with ANN (Salvetti+, 2017)
J/ApJS/235/39 : Jet properties of γ-ray-loud 3FGL AGNs (Chen, 2018)
J/other/RAA/18.56 : gamma-ray spectrum for Fermi blazars (Kang+, 2018)
J/ApJS/247/33 : The Fermi LAT 4th source catalog (4FGL) (Abdollahi+, 2020)
Byte-by-byte Description of file: table1.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 2 I2 --- Seq [1/34] Parameter label
4- 21 A18 --- Param Selected parameter
23- 26 F4.2 --- D-KS [0/0.76] Test statistic D for the two-sample
Kolmogorov-Smirnov test
28- 35 E8.3 --- p1-KS [0/1] p-value (p1) for the two-sample
Kolmogorov-Smirnov test
37- 42 F6.2 --- t-tt [-44/27] t-statistic (t) for the Welch
two sample t-test
44- 47 I4 --- df-tt [693/1767] Degree of freedom for the
t-statistic (df) for the Welch two sample t-test
49- 57 A9 --- p2-tt p-value(p2) for the Welch two sample t-test
59- 64 I6 --- W-W [51341/720456] Continuity correction test
statistic (W) for the Wilcoxon rank sum test
66- 74 A9 --- p3-W p-value (p3) for the Wilcoxon rank sum test
76- 80 F5.2 --- Gini [0.01/97.3] The Gini coefficient
--------------------------------------------------------------------------------
Byte-by-byte Description of file: table2.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 3 A3 --- CL Classifier
5- 6 I2 --- Num [8/17] Number of parameters for the optimal
combination
8- 10 I3 --- bll [812/923] Number of BL Lacs predicted by a
supervised classifier
12- 14 I3 --- fsrq [389/500] Number of FSRQs predicted by a
supervised classifier
16- 20 F5.3 --- Acc [0.918/0.93] Highest accuracies of each classifier
22- 22 I1 --- par1 [1/3] Label of parameter (see Table 1)
24- 25 I2 --- par2 [3/11] Label of parameter (see Table 1)
27- 28 I2 --- par3 [5/12] Label of parameter (see Table 1)
30- 31 I2 --- par4 [6/14] Label of parameter (see Table 1)
33- 34 I2 --- par5 [7/16] Label of parameter (see Table 1)
36- 37 I2 --- par6 [8/19] Label of parameter (see Table 1)
39- 40 I2 --- par7 [9/20] Label of parameter (see Table 1)
42- 43 I2 --- par8 [11/23] Label of parameter (see Table 1)
45- 46 I2 --- par9 [12/23]? Label of parameter (see Table 1)
48- 49 I2 --- par10 [14/23]? Label of parameter (see Table 1)
51- 52 I2 --- par11 [15/23]? Label of parameter (see Table 1)
54- 55 I2 --- par12 [17/23]? Label of parameter (see Table 1)
57- 58 I2 --- par13 [18/20]? Label of parameter (see Table 1)
60- 61 I2 --- par14 [19/21]? Label of parameter (see Table 1)
63- 64 I2 --- par15 [20/23]? Label of parameter (see Table 1)
66- 67 I2 --- par16 [22/23]? Label of parameter (see Table 1)
69- 70 I2 --- par17 [23/23]? Label of parameter (see Table 1)
--------------------------------------------------------------------------------
Byte-by-byte Description of file: table3.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 4 I4 --- Seq [1/1312] Sequential running identifier
6- 9 A4 --- --- [4FGL]
10- 22 A13 --- 4FGL 4FGL Source Name
24- 50 A27 --- Cntpt Counterpart name
52- 54 A3 --- C4FGL Optical class (always "bcu") (1)
56- 59 A4 --- CRF Random Forest classification
("bll": 835 occurrences or
"fsrq": 477 occurrences) (2)
61- 65 F5.3 --- PBi-RF [0/1] Probability BCU i belongs to BL Lac using RF
67- 71 F5.3 --- PFi-RF [0/1] Probability BCU i belongs to FSRQ using RF
73- 76 A4 --- CSVM Support Vector Machines classification
("bll": 858 occurrences or
"fsrq": 454 occurrences) (2)
78- 83 F6.4 --- PBi-SVM [0.016/1] Probability BCU i belongs to BL Lac
using SVM
85- 90 F6.4 --- PFi-SVM [0.0005/0.99] Probability BCU i belongs to FSRQ
using SVM
92- 95 A4 --- CANN Artificial neural network classification
("bll": 868 occurrences or
"fsrq": 444 occurrences) (2)
97-102 F6.4 --- PBi-ANN [0.1/0.9] Probability BCU i belongs to BL Lac
using ANN
104-109 F6.4 --- PFi-ANN [0.1/0.9] Probability BCU i belongs to FSRQ
using ANN
--------------------------------------------------------------------------------
Note (1): BCU as reported in The Fermi-LAT collaboration (2019, J/ApJS/247/33).
Note (2): Classification of the Fermi BCUs using the Random Forest (CRF),
Support Vector Machines (CSVM), and Artificial Neural Networks (CANN).
--------------------------------------------------------------------------------
History:
From electronic version of the journal
(End) Prepared by [AAS], Emmanuelle Perret [CDS] 28-May-2021