SEP forecasting: An ML approach utilizing solar flare and CME characteristics

<p>We present a novel analysis of historical events, i.e. Coronal Mass Ejections (CMEs) associated with solar flares resulting either in Solar Energetic Particle (SEP) events or not. We exploit a binary logistic regression model to construct an SEP forecasting model based on observed flare and CME characteristics. The logistic regression is applied within a machine learning (ML) framework, using a supervised ML model, which attempts to discriminate between two different binary classes/conditions (i.e. 0 in the case of no SEP and 1 in the case of an SEP event). The model is applied to separate cases of non-SEP events or SEP events (with two energy thresholds E>100MeV or E>300MeV). We utilize all available events to train the model and calculate the probability of an SEP event detection and also some representative metrics of its performance. The model was applied in two important Space Weather events namely July 23, 2012 and September 5, 2022 CMEs to examine the high-energy SEP production for these two extreme solar events. In both cases, we estimate a high probability (> 95%) of SEP events that were confirmed using STEREO-A and Solar Orbiter insitu data, respectively. The model is proven to be an excellent classifier for SEP identification by applying multivariate logistic regression.</p&gt