两种监督机器学习算法在Fermi BCU分类评估中的应用

Application of Two Supervised Machine Learning Classifiers in Classfication Evaluation of Fermi BCUs

在最新的费米第四期伽马射线源目录(4FGL)中,一共有3131个耀变体,包括1116个蝎虎天体(BLLacs),686个平谱射电类星体(FSRQs)和1329个未知类型的耀变体(BCUs).为了评估BCU可能的分类,通过双样本K-S检验选择合适参数,利用高斯混合有限模型(Mclust)和逻辑回归(LR)监督机器学习算法,对1329个BCUs的分类进行评估.评估可靠性检验结果表明,Mclust和LR两种算法的准确率分别为85.95%和89.46%.综合两种算法结果,给出了731个BLLacs和432个FSRQs候选体.

The new release of the fourth Fermi-LAT source catalog(4FGL) reported 3 131 blazar sources,including 1 116 BL Lacertaes(BL Lacs),686 flat-spectrum radio quasars(FSRQs) and 1 329 blazar candidates of uncertain type(BCUs).In order to evaluate the potential classification of BCUs,we use the two-sample KS test to select appropriate parameters.With the Gaussian mixture finite model(Mclust) and logistic regression(LR) classifier algorithms of supervising machine learning,we train and test the algorithm models with known samples and and then use it for classfication evaluation of unknown samples(BCUs).In the case where the overall accuracies of two classfiers are 85.95%and 89.46%,respectively,we obtain 731 BL Lac candidates and 432 FSRQ candidates.