Photometric redshift-aided classification using ensemble learning

We present SHEEP, a new machine learning approach to the classic problem of astronomical source classification, which combines outputs from XGBoost, LightGBM, and CatBoost algorithms create stronger classifiers. A novel step in our pipeline is that prior performing SHEEP first estimates photometric redshifts, are then placed into data set as an additional feature for classification model training; this results significant improvements subsequent performance. contains two distinct methodologies: (i) Multi-class (ii) one versus all with correction by meta-learner. demonstrate performance stars, galaxies, quasars using composed SDSS WISE photometry 3.5 million sources. The resulting F 1 -scores follows: 0.992 galaxies; 0.967 quasars; 0.985 stars. In terms 1-scores three classes, found outperform recent RandomForest-based essentially identical set. Our methodology also facilitates explainability via importances; it allows selection sources whose uncertain classifications may make them interesting follow-up observations.