Authors: Pouya Hosseinzadeh, Soukaina Filali Boubrahimi, Shah Muhammad Hamdi

Solar Energetic Particles (SEPs) are linked to severe solar events that have the potential to cause extensive damage to space-based and terrestrial life, as well as infrastructure. Among these events, high-intensity SEP events, specifically those with a 100 MeV energy level, pose a significant risk to Earth’s orbiting satellites, as they can expose astronauts to harmful radiation and lead to severe health issues. However, accurate prediction of these events is a major challenge due to the rarity of SEP events. To address this challenge, we aim to enhance the prediction of SEP events associated with 30 MeV, 60 MeV, and 100 MeV energy levels by artificially increasing the available SEP data samples. We explore the effectiveness of using univariate and multivariate time series data of proton flux as input for machine learning-based prediction methods. Our study focuses on solar cycles 22, 23, and 24. Our findings indicate that employing data augmentation techniques significantly improves the accuracy and F1-score of the aforementioned classifiers, particularly in the case of the Time Series Forest classifier.