Forecasting Solar Flares with Machine Learning Algorithms

Space Weather refers to the phenomena occurring on the Sun that affect Earth’s magnetosphere and ionosphere. These phenomena – among them solar flares – influence the performance and reliability of technological systems on Earth or in its near orbit. Since 2015, the HighPIDS research group has been dedicated to research on solar flares forecasting. To develop solar flares forecasting, the HighPIDS group researches strategies that use machine learning and deep learning algorithms, combined with methodologies for attribute selection, optimizing hyperparameters selection, unbalanced data handling, and finding a good relationship between forecast horizons and accuracy. The proofs-of-concept developed so far obtained accuracies between 70% and 85%, with forecast horizons from 24 to 72 hours in advance for high-intensity solar flares (≥ 𝐶 class). The proof-of-concept is available on the web as a service called Guaraci. This paper summarizes of the results achieved by the HighPIDS research group so far and some prospects for future research in solar flare forecasting.