Identification, mass and age of primary red clump stars from spectral features derived with the LAMOST DR7

TL;DR

This study develops a machine learning approach using XGBoost to accurately identify primary red clump stars, estimate their ages and masses, and analyze spectral features, achieving high purity and reliable distance measurements from LAMOST DR7 data.

Contribution

The paper introduces a novel application of XGBoost and SHAP for distinguishing RC and RGB stars, estimating stellar parameters, and validating feature importance with spectral data.

Findings

Achieved over 90% purity in identifying primary RC stars.

Estimated stellar ages and masses with uncertainties of 31% and 13%.

Demonstrated effectiveness at lower spectral resolution (R~250).

Abstract

Although red clump (RC) stars are easy to identify due to their stability of luminosity and color, about 20-50% are actually red giant branch (RGB) stars in the same location on the HR diagram. In this paper, a sample of 210,504 spectra for 184 318 primary RC (PRC) stars from the LAMOST DR7 is identified, which has a purity of higher than 90 percent. The RC and the RGB stars are successfully distinguished through LAMOST spectra(R~1800 and SNR>10) by adopting the XGBoost ensemble learning algorithm, and the secondary RC stars are also removed. The SHapley Additive exPlanations (SHAP) value is used to explain the top features that the XGBoost model selected. The features are around Fe5270, MgH & MgIb, Fe4957, Fe4207, Cr5208, and CN, which can successfully distinguish RGB and RC stars. The XGBoost is also used to estimate the ages and masses of PRC stars by training their spectra with Kepler labeled asteroseismic parameters. The uncertainties of mass and age are 13 and 31 percent, respectively. Verifying the feature attribution model, we find the age-sensitive elements XGBoost gets are consistent with the literature. Distances of the PRC stars are derived by $K_{S}$ absolute magnitude calibrated by Gaia EDR3, which has an uncertainty of about 6 percent and shows the stars mainly locate at the Galactic disk. We also test the XGBoost with R$\sim$250, which is the resolution of the Chinese Space Station Telescope(CSST) under construction, it is still capable of finding sensitive features to distinguish RC and RGB.