Feature Selection for Geomagnetic Storm Forecasting Using Particle Swarm Optimisation

Geomagnetic activity is a significant research area due to its impact on navigation, satellites, and other technologies. The geomagnetic index Kp is particularly essential for monitoring such activity. Forecasting geomagnetic storms can provide early warnings, helping to mitigate potential disasters. Various methods, including empirical models, physics-based models, and machine learning models, are employed for forecasting. This study focuses on Machine learning models, emphasising selecting appropriate features to enhance prediction accuracy. The primary aim was to improve the 3-hour Kp predictions compared to prior research. Model performance was primarily evaluated using the R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> metric. Additionally, RMSE was calculated to ensure comparability with studies that did not use R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>. The model’s ability to detect storms was assessed by classifying results as Kp>5 (storm) or Kp<5 (not a storm), with the F1 score used as the evaluation metric for classification. The best model achieved an R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.71, an RMSE of 0.6671, and an F1 score of 0.6.