Machine learning to predict star formation rates and stellar masses from photometric data of galaxies Agharini Linda Ariyani (a*), Mochamad Ikbal Arifyanto (a)
Department of Astronomy, Institut Teknologi Bandung, Bandung, West Java, Indonesia
*agharini2906[at]gmail.com
Abstract
Stellar mass (SM) and star formation rate (SFR) are key diagnostics of galaxy evolution, yet their estimation through spectral energy distribution (SED) fitting is computationally demanding and impractical for large surveys. In this study, we evaluate machine learning (ML) methods as efficient alternatives for deriving these properties from photometric data. XGBoost---which showed the best performance among other ML architectures---were trained and tested on photometry data from the GAMA Panchromatic Data Release as features and MAGPHYS-derived properties as targets. We achieved RMSE values of 0.073 dex (SM) and 0.160 dex (SFR). SM predictions were consistently more accurate than SFR, reflecting the greater variability and dust dependence of star formation activity. Compared to traditional SED fitting, the ML models replicate results with competitive accuracy while reducing computation time by several orders of magnitude. These findings highlight the potential of ML to enable fast and scalable galaxy property estimation in upcoming large-scale surveys such as LSST.
