Abstract:
Since the discovery of the first exoplanet in 1995 orbiting a Sun-like star, more than 5,867 exoplanets have been identified, according to the NASA Exoplanet Archive. Knowing that planetary parameters depend directly on stellar parameters, this work aimed to improve the characterization of exoplanet host stars observed by the J-PLUS and S-PLUS photometric surveys using machine learning techniques, with a focus on the Random Forest and XGBoost algorithms. The training of the models used photometric data from the 12 optical filters of J-PLUS and S-PLUS combined with data from the LAMOST and APOGEE surveys to predict stellar parameters such as effective temperature (Teff), surface gravity (log g), and metallicity ([Fe/H]). The models were applied to host stars listed in the catalogs of the Kepler and TESS missions and the HARPS spectrograph. The results highlighted the effectiveness of the Random Forest algorithm, which showed greater accuracy than XGBoost, particularly for Teff and log g. The significant reduction in errors, when compared to literature values, reaffirms the viability of combining machine learning with a photometric system that integrates broad, intermediate, and narrow-band filters to obtain precise estimates of stellar parameters. In addition to predicting these parameters, fundamental stellar properties such as luminosity, absolute magnitude, bolometric magnitude, mass, and radius were derived. These results were used to characterize exoplanets with transit values available in the Kepler and TESS catalogs in fields overlapping with J-PLUS and S-PLUS. This study demonstrates that machine learning techniques not only enhance the characterization of host stars but also open new possibilities for identifying exoplanets and other objects of interest, such as eclipsing binaries and brown dwarfs. Finally, the findings are promising for future analyses, especially with the application of this methodology to other surveys like J-PAS, which features a larger number of filters and the potential for even more accurate results.
