A Comprehensive Analysis of Machine Learning Approaches for Solar Flare Prediction

Solar flares have a severe impact on near-Earth space weather due to the release of energy in the form of electromagnetic radiation bursts and energetic particles. They pose serious threats to satellites, astronaut health, communication systems, and other space reliant technologies. Therefore, forecasting solar flare has a growing demand in the space sciences. Recent advances in machine learning allow handling of high dimensional data and have shown potential in flare forecasting. Previous studies have shown that solar active region (AR) properties can be well constrained by several magnetic field parameters deduced from vector magnetogram data. We conduct a comparative analysis of four robust Machine learning models to classify the ARs based on their flaring capability by training them on magnetic features derived from Spaceweather HMI Active Region Patch (SHARP) data. We demonstrate that Logistic Regression and Support vector Machine performs extremely well achieving the highest True Skill Score. In addition, we perform a correlation analysis on the input magnetic features and rank them according to their ability in predicting AR flare productivity. Our study sheds light on the most promising machine learning algorithms for solar flare forecasting and indicates which AR parameters play an important role in achieving a high prediction skill.