CAOSP abstracts, Volume: 55, No.: 3, year: 2025

Author(s): Parimucha, Š.; Gajdoš, P.; Markus, Y.; Kudak, V.
Journal: Contributions of the Astronomical Observatory Skalnaté Pleso, vol. 55, no. 3, p. 370-373.
Date: 04/2025
Title: Prediction of photometric parameters of overcontact eclipsing binaries using deep-learning model
Keyword(s): eclipsing binaries, machine-learning, deep-learning
Pages: 370 -- 373
DOI: https://doi.org/10.31577/caosp.2025.55.3.370
NASA ADS Abstract Service: 2025CoSka..55c.370P
Citation:
License: This work is licensed under a Creative Commons Attribution 4.0 International License

Abstract: This paper presents a deep-learning model for predicting the photometric parameters of W UMa-type eclipsing binaries without spots. The model assumes that the light curve (LCs) can be described by four parameters: the orbital inclination, photometric mass ratio, temperature ratio, and common potential. The training dataset comprises 500~000 simulated LCs in the Gaia G passband. The best results were obtained using the random forest predictor, achieving a Mean Absolute Percentage Error (MAPE) of 6.1%. The study concluded that the quality of parameter prediction strongly depends on the quality of the analyzed LCs curves, requiring careful data preprocessing.