Evaluating the Geoeffectiveness of Interplanetary Coronal Mass Ejections: Insights from a Support Vector Machine Approach with SHAP Value Analysis

Abstract

Abstract In this study, we compiled a data set of 510 interplanetary coronal mass ejections (ICME) events from 1996–2023 and trained a radial basis function support vector machine (RBF-SVM) model to investigate the geoeffectiveness of ICMEs and its dependence on the solar wind conditions observed at 1 au. The model demonstrates high performance in classifying geomagnetic storm intensities at specific Disturbance Storm Time thresholds and evaluating the geoeffectiveness of ICMEs. The model’s output was assessed using precision, recall, F1 score, and true skill statistics (TSS), complemented by stratified k-folds cross-validation for robustness. At the −50 nT threshold, the model achieves precisions of 0.84 and 0.93, recalls of 0.94 and 0.82, and corresponding F1 scores of 0.89 and 0.87 for the categories separated by this threshold, respectively. Overall accuracy is noted at 0.88, with a TSS of 0.76. Despite challenges at the −100 nT threshold due to data set imbalance and limited samples, the model maintains an overall accuracy of 0.87, with a TSS of 0.69, demonstrating the model’s ability to effectively handle imbalanced data. Physical insights were gained through model explanation with a SHapley Additive exPlanations (SHAP) value analysis, pinpointing the role of the southward magnetic field component in triggering geomagnetic storms, as well as the critical impact of shock-ICME combinations in intensifying these storms. The effective application of an SVM model with SHAP value analysis offers a way to understand and predict the geoeffectiveness of ICMEs. It also underscores the capability of a relatively simple machine learning model in predicting space weather and revealing the underlying physical mechanisms.