Calibrating machine learning with multi-band photometry: Resolving parameter degeneracies in contact binary NSVS 4803568

Xu-Zhi Li, Xue-Tong Liu, Wei Liu, Xu-Dong Zhang,
Calibrating machine learning with multi-band photometry: Resolving parameter degeneracies in contact binary NSVS 4803568,
New Astronomy,
2025,
102484,
ISSN 1384-1076,
https://doi.org/10.1016/j.newast.2025.102484.
(https://www.sciencedirect.com/science/article/pii/S1384107625001344)
Abstract: Contact binary stars are crucial for studying stellar evolution and merger events, but precise determination of their physical parameters (mass ratio q, inclination i, fill-out factor f) is challenging. While large-scale, single-band surveys coupled with machine learning enable rapid population studies, discrepancies arise between solutions derived from such automated pipelines due to inherent degeneracies and limited wavelength constraints. In this work, we resolve the conflicting parameters reported for contact binary system NSVS 4803568 through comprehensive multi-band photometric observations. We conducted follow-up B, V, R, and I-band observations and analyzed the light curves using PHOEBE, refined via Markov Chain Monte Carlo. Our solution confirms the system as a W-subtype contact binary and reveals a significant third-light contribution. The parameter inconsistencies between single-band studies highlight limitations in machine learning training data sensitivity and the need for multi-dimensional flux constraints. We advocate a hybrid approach: machine learning pre-screening of survey data to identify high-priority targets (e.g., extreme q or short period systems) followed by multi-band validation to calibrate models and mitigate systematic errors. This synergy is essential for unlocking the full potential of astronomical big data in stellar astrophysics.
Keywords: Eclipsing binary; Contact binary; NSVS 4803568