Classication for BZUs in 5BZCAT

In order to evaluate the potential optical classification of 227 BZUs in 5BZCAT, we divide the BZUs into BL lac candidates and FSRQ candidates by four machine learning methods: support vector machine (SVM), random forest (RF), ensemble learning (EM) and multi-layer perceptron (MLP). And the classification accuracy is improved by feature engineering and grid search. By combining the classification results of four classifiers and setting the threshold of discrimination probability to 0.8, we get 33 BL lacs candidates and 119 FSRQs candidates.