Authors: Sanchita Pal (NASA GSFC), Thomas Narock (Center for Natural, Computer, and Data Sciences, Goucher College), Luiz F. G. dos Santos (SETI Institute), Andreas J. Weiss (NASA GSFC), Ayris Narock (NASA GSFC and ADNET Systems Inc), Lan K. Jian (NASA GSFC), Teresa Nieves-Chinchilla (NASA GSFC), Simon W. Good (University of Helsinki, Finland)

Solar wind in the interplanetary medium may originate from different solar sources, e.g., coronal holes, solar transients like coronal mass ejections (CMEs), and streamer belt regions based on which the solar wind may be categorized. We introduce techniques that use supervised machine learning (ML) models to classify solar wind based on their magnetic and plasma characteristics. We propose two independent pipelines where the first comprises a convolutional neural network and a support vector machine and flags the solar wind stream whenever geoeffective transients like interplanetary (I)CMEs are present. The second pipeline is a probabilistic neural network to categorize solar wind with uncertainty based on their magnetic and plasma characteristics. While validating with real data, the first pipeline that works with only solar wind magnetic characteristics has an accuracy of 88% and the second model has an accuracy of 96%. Both the models can be utilized in real-time solar wind and offer warnings of disrupted space weather, which could lower possible risks associated with space weather disturbances.