Abstract:

Asteroseismology, the study of stellar pulsations, provides great insight into a star’s physical parameters, and with different computational methods, we are able to estimate the unknown characteristics of stars (like chemistry or age). In this project, we present a web-based interface with the ability to predict the physical properties of solar-like stars from stellar observations with the use of machine learning. With perturbation techniques and a Classification and Regressions Trees Algorithm (CART) on 350,000 theoretical stellar models to learn the matching between unknown and known properties of the star and predict its unknown parameters with a trained regression model. Initially developed with a database, the architecture of the interface consists of two separate frame-works: a back-end model and a dynamic front-end web application. With an additionally implemented, asynchronous front-end execution, email delivery server, and online database for storage of data and results, the interface successfully handles incoming and existing users and data. Finally, the mean predictions of 16 Cygni A, 16 Cygni B, and KIC 12258514 resulted in a 3.11% difference to original predictions, and the average prediction time was 6.27 ±1.06 seconds per star. This novel engineering project is potentially another great tool for stellar research which we plan to deploy simultaneously with the submission of a future publication that extends the back-end algorithm to handle more kinds of stars.