Prediction of Fundamental Stellar Parameters of F Type Star- A Comprehensive Study using Bayesian Neural Network.

Accurate determination of stellar parameters is crucial for understandingstellar evolution, exoplanet habitability, and chemical peculiarities in stellaratmospheres. Our work aims to improve the precision of predicting stellarproperties—effective temperature (Teff), metallicity ([Fe/H]), and surfacegravity (log g)—by leveraging advanced machine learning techniques. [1]We employ a medium-density neural network with a non-linear activationfunction, trained using a dataset of 588 F-type stars selected from theXSHOOTER, [2] ELODIE 3.0, MILES, and Indo-US stellar libraries. [3]Focusing on hydrogen Balmer lines and metallic lines in the visible range(390–550 nm), as given by the LICK/ADS database, [4] we reprocess stellarspectra through pseudo-continuum normalization, equivalent widthmeasurements, and model fitting using the SPECTRUM code within theiSpec library. [5] Given the limited and potentially noisy nature of thedataset, Bayesian regularization [6] is implemented to prevent overfittingand enhance generalization. Our study also investigates the spectralevolution of F-type stars across their population distribution—spanningdwarfs, subgiants, and giants—providing insight into how key spectralfeatures change with stellar age and composition. Since F-type stars sharemany properties with G- and K-type stars, which are known to hosthabitable zones, a refined understanding of their stellar characteristics isessential for exoplanet studies. Moreover, our model can identify chemicallypeculiar F-type stars, [7] particularly those in pre-main-sequence stages, bycorrelating fundamental parameters with spectral line strengths. Byimproving the precision of stellar parameter determination, our workcontributes to a deeper understanding of F-type stars, their potential forhosting habitable planets, and the broader implications for stellarastrophysics