Revisiting Fourier-Based Metallicity Estimations of RR Lyrae Light Curves Using Machine Learning

Orem, UT

Decades of research have established that the metallicity [Fe/H] of RR Lyrae variable stars can be estimated by analyzing the shape of their light curves through Fourier decomposition (Jurcsik and Kovács 1996). They achieved this using multiple linear regression and found a strong relationship between the star’s pulsation period and a phase parameter called φ31. In this study, we investigate whether modern machine learning models (ML) can improve the accuracy of these estimates. We confirmed the original 1996 relation using its dataset of 81 stars as a baseline. For this work, we applied a selection of ML models from the scikit-learn Python library, including Lasso, Ridge, and ElasticNet and found that they did outperform the original relation. Both the original relation and our ML models were applied to more recent RR Lyrae data to explore their efficacy across multiple datasets.