Estimation of Stellar Atmospheric Parameters with Light Gradient Boosting Machine Algorithm and Principal Component Analysis

Abstract

Abstract In this paper, we propose a new method to estimate stellar atmospheric parameters with photometric data, which is based on principal component analysis (PCA) and light gradient boosting machine (LightGBM) algorithms. We first use PCA to transform all band photometric data (u, v, g, r, i, and z) and then utilize LightGBM to estimate stellar atmospheric parameters. The experimental results show that the root mean square errors of the method for estimating the effective temperature, surface gravity, and metallicity are 90 K, 0.40 dex, and 0.20 dex, respectively. We then compare PCA + LightGBM with the original photometry data (OPD) + LightGBM and the color index data (CID) + LightGBM. The experimental results show that the performance of PCA + LightGBM is better than that of CID + LightGBM and OPD + LightGBM, and PCA + LightGBM can solve the problems of model instability and inaccurate estimation results caused by direct use of OPD or CID as input for LightGBM. We believe the new features obtained by PCA can be used on photometric data collected by the Chinese Space Station Telescope.