Physical Predictors of Halo Coronal Mass Ejection Formation Identified via Explainable Machine Learning: Linking Active Region Evolution to Solar Flares and Type II Radio Bursts

Solar flares, coronal mass ejections (CMEs), and solar radio bursts are coupled drivers of the most energetic space-weather events, with halo CMEs of particular operational importance for Earth-directed CMEs and intense Type II radio burst activity. This study applies explainable AI, SHapley Additive exPlanations (SHAP) analysis of a random forest classifier, to identify the photospheric magnetic field parameters that most strongly govern whether a CME accompanying a flare-associated Type II burst attains halo morphology. Time-series features are derived from Solar Dynamics Observatory/Helioseismic and Magnetic Imager (SDO/HMI) Space-weather HMI Active Region Patches (SHARP) data spanning up to 24 hr before Type II burst onset. To address nonhalo underrepresentation, three augmentation methods (scaling, jitter, and window slicing) are applied under a leakage-free protocol, and classifiers are trained across 48 dataset variants spanning 24 observation windows. Augmentation improves the F1 score and true skill statistic (TSS) in all 48 experiments, with a representative model (11 hr window, normalized features) achieving F1 = 0.893 and TSS = 0.781. Across 100 repeated train/test splits the discriminative ability is stable (ROC-AUC = 0.987 ± 0.013), with threshold-dependent metrics showing wider spread at this sample size (mean F1 = 0.761). SHAP analysis identifies current helicity ( MEANJZH ), the twist parameter ( MEANALP ), and the rate of change of magnetic shear and inclination ( MEANSHR , MEANGAM ) as the leading predictors, stable across all folds. These quantities trace the build-up of a twisted, current-carrying flux rope toward the Kruskal–Shafranov kink-instability threshold, linking active-region evolution, flare energetics, and the strength of the CME-driven shock that excites the Type II emission.

Authors

Institutions

Publication Details

Journal
The Astrophysical Journal Supplement Series
Published
2026-11-18
DOI
https://doi.org/10.3847/1538-4365/aeb2ff
Primary Topic
Solar and Space Plasma Dynamics
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Physical Predictors of Halo Coronal Mass Ejection Formation Identified via Explainable Machine Learning: Linking Active Region Evolution to Solar Flares and Type II Radio Bursts

Abd-ur Raheem, Hüseyin Çavuş, Haimin Wang, Jason T. L. Wang et al.
1 links
The Astrophysical Journal Supplement Series
Solar and Space Plasma Dynamics
article

Physical Predictors of Halo Coronal Mass Ejection Formation Identified via Explainable Machine Learning: Linking Active Region Evolution to Solar Flares and Type II Radio Bursts

Abd-ur Raheem, Hüseyin Çavuş, Haimin Wang, Jason T. L. Wang, Gani Caglar Coban
article en

Abstract

Solar flares, coronal mass ejections (CMEs), and solar radio bursts are coupled drivers of the most energetic space-weather events, with halo CMEs of particular operational importance for Earth-directed CMEs and intense Type II radio burst activity. This study applies explainable AI, SHapley Additive exPlanations (SHAP) analysis of a random forest classifier, to identify the photospheric magnetic field parameters that most strongly govern whether a CME accompanying a flare-associated Type II burst attains halo morphology. Time-series features are derived from Solar Dynamics Observatory/Helioseismic and Magnetic Imager (SDO/HMI) Space-weather HMI Active Region Patches (SHARP) data spanning up to 24 hr before Type II burst onset. To address nonhalo underrepresentation, three augmentation methods (scaling, jitter, and window slicing) are applied under a leakage-free protocol, and classifiers are trained across 48 dataset variants spanning 24 observation windows. Augmentation improves the F1 score and true skill statistic (TSS) in all 48 experiments, with a representative model (11 hr window, normalized features) achieving F1 = 0.893 and TSS = 0.781. Across 100 repeated train/test splits the discriminative ability is stable (ROC-AUC = 0.987 ± 0.013), with threshold-dependent metrics showing wider spread at this sample size (mean F1 = 0.761). SHAP analysis identifies current helicity ( MEANJZH ), the twist parameter ( MEANALP ), and the rate of change of magnetic shear and inclination ( MEANSHR , MEANGAM ) as the leading predictors, stable across all folds. These quantities trace the build-up of a twisted, current-carrying flux rope toward the Kruskal–Shafranov kink-instability threshold, linking active-region evolution, flare energetics, and the strength of the CME-driven shock that excites the Type II emission.

The Astrophysical Journal Supplement SeriesVol. 287(2)
Çanakkale Onsekiz Mart Üniversitesi (TR), New Jersey Institute of Technology (US), University Heights Center (US)
National Aeronautics and Space Administration
Solar and Space Plasma Dynamics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon