CME arrival modeling with machine learning

Space weather phenomena have long captured the attention of the scientific community, and along with recent technological developments, the awareness that such phenomena can interfere with human activities on Earth has grown considerably. Coronal mass ejections (CMEs) are among the main drivers of space weather. Therefore, developing tools to provide information on their arrival at Earth's nearby space has become increasingly important. Liu et al. developed a tool, called CME Arrival Time Prediction Using Machine Learning Algorithms (CAT-PUMA), to obtain fast and accurate predictions of CME transit time. This present work aims at the expansion of the CAT-PUMA concept, employing supervised learning to obtain vital information about the arrival of CMEs at Earth. In this study, we report the results of our work following the implementation of supervised regression and classification models in the CAT-PUMA framework. We conducted a comparison of various machine learning models in the context of predicting the transit time of CMEs and classifying CMEs as either Earth impacting or non-impacting. In this way, we are able to provide information on the possibility of a CME reaching Earth relying on CME features and solar wind parameters measured at take-off. This application thus provides quantitative indications about the geoeffectiveness of these space weather events. While machine-learning models can demonstrate fairly strong performance in regression and classification tasks, it is not always straightforward to extrapolate their practical potential and real-world applicability. To address this challenge, we employed model interpretation techniques, specifically Shap values, to gain quantitative insights into the limitations that affect these models.