Machine-learning analysis of solar flare light-curve morphology and implications for stellar CME prediction

ABSTRACT Coronal mass ejections (CMEs) are major drivers of stellar space weather and can influence exoplanet habitability, yet confirmed stellar CMEs remain extremely rare compared to the frequent occurrence of white-light flares. Whether flares are commonly accompanied by CMEs is a key question in solar–stellar comparative studies. We compiled a sample of 1156 M-class and larger solar flares observed by the Extreme ultraviolet SpectroPhotometer (ESP; 0.1–7 nm), part of the Extreme Ultraviolet Variability Experiment (EVE) onboard the Solar Dynamics Observatory (SDO), and extracted flare features using convolutional neural networks and manual methods. Six machine-learning classifiers were trained to distinguish eruptive from confined events, with logistic regression achieving the best performance (true skill statistic; TSS = 0.24), indicating moderate predictive capability. Normalized white-light and ESP broad-band light curves show strong correlations, suggesting that the model can be transferred to stellar flare observations. Applying the model to 14 979 superflares on FGKM-type main-sequence stars and 2738 on Sun-like stars observed by Kepler, we find that $\sim$44 per cent and $\sim$49 per cent of these events, respectively, are likely CME-associated. The corresponding fractions for F, G, K and M stars are $\sim$38 per cent, $\sim$45 per cent, $\sim$47 per cent, and $\sim$40 per cent. These results indicate that fewer than half of stellar superflares are accompanied by CMEs, likely reflecting the strong background magnetic fields of their host stars that can confine eruptions. This study provides a systematic exploration of CME occurrence in stellar superflares and carries important implications for stellar activity, space weather, and the habitability of exoplanets.