Solar energetic particles (SEPs) are one of the most crucial aspects of space weather, predominantly constituting intense proton beams from the Sun. Those surpassing the Space Weather Prediction Center’s S1 threshold of a solar radiation storm have severe technological and biological implications for space missions outside the Earth’s magnetic field. Predicting their arrival in near-Earth space strongly depends on various factors, including the solar eruptions, such as flares and coronal mass ejections (CMEs). Therefore, we follow a strategic path toward SEP forecasting by leveraging machine learning (ML) principles for ‘research to operations’ demands. For efficient applications, there are two aspects in this dissertation: (1) the development of a benchmark data set of SEP events and (2) the design of a robust methodology for short-term predictions of SEPs using time-series-based ML classifiers.

The first outcome of this research is the Geostationary Solar Energetic Particle (GSEP) events data set, a comprehensive collection of over 400 SEP events spanning solar cycles 22 - 24. Each event in this data set has been meticulously analyzed for its spatiotemporal properties, including the characteristics of the associated source eruptions, such as flare magnitudes, locations, rise times, speeds and widths of CMEs, and radio bursts.

Next, we present an ensemble learning approach that merges the results from univariate time series of solar protons and X-ray fluxes. Furthermore, we demonstrate a proof-of-concept of using feature-based classifiers to distinguish SEP events from non-events. Here, non-events constitute those SEPs below the S1 threshold and “SEP-quiet” periods. We do a comparative analysis of the models and extensively evaluate the feasibility of our data-driven approach by establishing confidence in the predictions utilizing multiple forecasting metrics. Lastly, we present our results for lead times of up to 60 minutes that display great potential for the implementation of our methodology in near-real-time operations for short-term SEP event predictions.