Alexis J. Wu, Canyon Crest Academy; Yunxiang Liu, University of Colorado Boulder, Boulder

View Abstract Sign in for premium content


This paper introduces a machine learning approach to investigate the feature importance for scintillation forecasting. Here, the features are historical measurements used as input for machine learning models. We propose to use gradient boosting as the machine learning algorithm to conduct a scintillation forecasting task at high latitudes. Once the gradient boosting model is trained, the rank of feature importance can be obtained. The preliminary results show that the top 10 most important features indeed are correlated with the future occurrence of scintillation. The feature importance ranking has the potential to guide feature selection for machine learning-based scintillation forecasting and improve forecasting performance. In addition, the feature importance list could also provide insights on the investigation of the complex coupling between solar wind and ionospheric disturbance.