Solar Flare Classification using Modified Metaheuristic Optimized XGBoost

The Sun has a significant effect on nearly everything in the solar system. While a relatively tame star, it is nevertheless prone to strong emissions of energy in the form of solar flares. Events frequency follows a cycle of approximately 11 years. At the peak of this cycle, events occur more often. Powerful events can even reach the Earth and cause magnetic storms that can disrupt radio communications, satellite trajectories, and navigation. Particular energetic events can even cause significant harm to power and telecommunications infrastructure. A need for a robust system capable of accurately handling classification is evident. This work proposes an approach based on extreme gradient boosting optimized via metaheuristic algorithms and applied to a real-world observation dataset. Outcomes have been compared to several modern optimization algorithms and the outcomes are promising. The results have been statistically validated as significant, and the best models analyzed to determine feature impacts on classifications.