Advanced Classification of Hot Subdwarf Binaries Using Artificial Intelligence Techniques and Gaia DR3 data
Hot subdwarfs are compact blue evolved objects, burning helium in their cores surrounded by a tiny hydrogen envelope. Most models agree on a common envelope binary evolution scenario in the Red Giant phase. However, the binarity rate for these objects is yet unsolved. We aim to develop a novel classification method for identifying hot subdwarf binaries within large datasets using Artificial Intelligence methods and Gaia DR3 data. The results will be compared with those obtained previously using VOSA (Virtual Observatory Sed Analyzer) on coincident samples. The methods include several machine learning techniques. We used Support Vector Machines (SVM) to classify 3084 hot subdwarf stars based on their color-magnitude properties. Of these, 2815 objects have Gaia Data Release 3 BP/RP spectra, which were classified using Self-Organizing Maps (SOM) and Convolutional Neural Networks (CNN). The findings demonstrate a high agreement level (70-90%) with VOSA's classification, indicating that machine learning methods effectively classify sources with an accuracy comparable to human inspection or non-AI techniques. SVM in a radial basis function achieves 70.97% reproducibility for binary targets using photometry. CNN reaches 84.94% for binary detection using spectroscopy. We also found that the single-binary differences are especially observable on the infrared flux in our GDR3 BP/BR spectra, at wavelengths larger than 700 nm. We found that all our methods are effective in discerning between single and binary systems and are consistent with the results previously obtained with VOSA. In global terms, considering all quality metrics, CNN is the method that provides the best accuracy. The methods are also effective for detecting peculiarities in the spectra. Further research is needed to refine our techniques and enhance automated classification reliability, especially for large-scale surveys.