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

Abstract

Hot subdwarf stars are very important for understanding stellar evolution, stellar astrophysics, and binary star systems. Identifying more such stars can help us better understand their statistical distribution, properties, and evolution. In this paper, we present a new method to search for hot subdwarf stars in photometric data (BP, RP, G, g, r, i, z, y) using a machine-learning algorithm, a graph neural network, and a Gaussian mixture model. We use a Gaussian mixture model and Markov distance to build the graph structure, and on the graph structure we use a graph neural network to identify hot subdwarf stars from a dataset containing 31838 stars, with the recall, precision, and F1 score maximized on the original, weight, and synthetic minority oversampling technique datasets. Finally, to validate the model, we selected about 2116 hot subdwarf candidates from the Gaia Data Release 3 database and compared them with the studies by Culpan et al. (2022, A&A, 662, A40) and Geier et al. (2019, A&A, 621, A38).