Effective temperatures estimation of low-mass stars and identification of T Tauri stars in LAMOST DR10 using machine learning

Abstract We present effective temperature (Teff) estimates of low-mass stars and T Tauri stars (TTS) candidate detections derived from automated spectroscopic measurements and low-complexity machine-learning models applied to the LAMOST DR10 V2 survey. Equivalent widths of key diagnostic spectral features, including the atomic lines Hα, Li i λ6708Å and TiO/VO molecular bands, are automatically measured for all stars within 1 kpc observed by LAMOST. Using nine spectral features as inputs, we train a Gradient Boosting Machine tree-based regression model, calibrated with synthetic spectra from the PHOENIX library, to predict Teff over the range 2, 500 − 5, 100 K. We apply a logistic regression model to the principal components derived from the measured spectral features, enabling efficient identification of TTS candidates. Both models exhibit strong performance in validation tests. Finally, a Monte Carlo framework is employed to propagate input uncertainties and estimate Teff and its associated uncertainties for low-mass stars from 1,733,802 spectra and to identify 2,534 candidate TTS from 3,121 spectra.