PREDNASAJUCI / LECTURER :

Andrii Maliuk

(1) Astronomical Institute, Slovak Academy of Sciences,
059 60 Tatranská Lomnica

NAZOV / TITLE :

From Light Curves to Stellar Parameters: Machine Learning Applications for
Contact Binaries

ABSTRAKT / ABSTRACT :

Contact binary stars are key laboratories for studying stellar structure and
evolution, yet determining their fundamental parameters from light curves
remains computationally expensive due to the nonlinear nature of the inverse
problem. In this work, we present a machine learning (ML) framework for
estimation of primary physical parameters of contact binaries, including mass
ratio, fill-out factor, orbital inclination, and component temperatures.
Using a large synthetic dataset of light curves generated with the PHOEBE code,
we trained an ensemble ML model combining Random Forest, XGBoost, and CatBoost
with stacking regression. The model demonstrates robust performance on synthetic
test sets, achieving low root-mean-square errors across all target parameters.
Application to TESS-observed contact binaries shows that ML-predicted parameters
produce synthetic light curves that closely match observations, often
outperforming catalog-based solutions in chi-square statistics. Our results
highlight the potential of ML as a rapid and accurate complementary tool for
large-scale analysis of contact binaries, reducing computational cost by orders
of magnitude compared to traditional forward-modeling approaches.