Identifying hot subdwarf stars from photometric data using a Gaussian mixture model and graph neural network.

This article focuses on developing a novel method to identify hot subdwarf stars—small-mass, core helium-burning stars important for stellar evolution studies—using photometric data and advanced machine-learning techniques. The authors apply a graph neural network (GNN) model called GraphSAGE, combined with a Gaussian mixture model (GMM) to construct a graph based on stellar photometric features and Markov distance, addressing the challenge of imbalanced datasets through the synthetic minority oversampling technique (SMOTE). Experiments demonstrate that GraphSAGE outperforms other GNN variants and traditional machine-learning algorithms in classifying hot subdwarf stars, achieving high precision and recall. Applying this method to Gaia Data Release 3 and PanStarrs DR1 data, the study identifies 3,542 candidate hot subdwarf stars, with results showing improved accuracy and the ability to detect candidates beyond conventional selection criteria.