Applying Network Analysis and Feature Engineering in Machine Learning Models for Solar Forecasting

Bachelor's thesis at University of Colorado Boulder.

Solar flare and Coronal Mass Ejection (CME) prediction helps us protect delicate systems on Earth and in space. The causes of solar storms are not understood well enough to predict solar flares from preceding factors, so researchers are using advanced machine learning techniques to learn complex interactions and accurately predict solar flares. Part of these advanced techniques is synthesizing data from instrument readings to improve machine learning models using
feature engineering.

In my thesis, I focused on full-disk analysis to create novel features for use in machine learning models to predict solar flares. I applied network analysis to explore the ways active regions influence each other and to analyze the efficacy of using full-disk features for solar prediction. Through this analysis, I computed features and compared their effectiveness both using statistical tests and by training basic neural networks.