Abstract - Yihong Song, Rohit Prasanna

Prediction of Stellar Activity through Machine Learning and Asteroseismic Analysis

This project presents a novel technique to predict solar magnetic activity of distant stars through asteroseismic analysis and machine learning algorithms. Asteroseismology, the study of “star”quakes, provides astrophysicists with accurate information about stellar mass, radius, and age. This information may be derived from a Fourier transformation on the time series data of either star surface velocity or star luminosity, both expressions of energy flow. The resultant frequency spectra contain valuable information about the star and were previously analyzed through purely theoretical methods. However, with access to advanced tools such as machine learning, these data may now be effectively analyzed and may reveal hidden correlations. Our method utilizes a neural network trained on the Sun’s light curve (Luminosity Oscillations Imager) to predict the Sun's magnetic activity trend indicated by sunspot number (Sunspot Index and Long-term Solar Observations) for solar cycle 23. The model was then cross validated for accuracy with the sun's surface velocity data (Birmingham Solar Oscillations Network) for both solar cycles 23 and 24, where it scored a mean squared error of 0.16. The prediction showed significant visual correlations with the observed indicators. The ability of the model to predict accurately on both surface velocity and light curve data serves as justification for applying the technique to light curves of Kepler stars. Stellar activity data plays a major role in determining exoplanet habitability and surface conditions. The use of a machine learning model improves prediction accuracy and provides scalability when compared to present methods.