An Improved Halo Coronal Mass Ejection Geoeffectiveness Prediction Model Using Multiple Coronal Mass Ejection Features Based on the DC-PCA-KNN Method

Abstract

Coronal mass ejections (CME) are regarded as the main drivers of geomagnetic storms (GSs). In the prediction of geoeffectiveness, various CME features have been introduced without adequately considering the geoeffectiveness of CMEs and strong correlations among the features. In this study, a feature dimension reduction method combining distance correlation (DC) and principal component analysis (PCA) was employed for the K-nearest neighbors (KNN) model to predict the geoeffectiveness of halo CME by using the multiple CME features. First, based on CME features and the Disturbance Storm Time index, there are 169 CME-related GS (CME-GS) pairs that were defined as positive samples during the entire phases of solar cycles 23 and 24. Next, DC was used to screen the eight original CME features. It was found that the three CME features, such as acceleration, kinetic energy, and measurement position angle, were weakly correlated with CME-GS pairs according to the performance of KNN prediction models and then were discarded. In order to further reduce the dimension of the input features for the KNN prediction model, PCA was subsequently applied. And the above remaining five principal components were reduced to one. Finally, the DC-PCA-KNN prediction model, achieving a mean true positive rate of 0.6519 and an accuracy of 0.6494, performed well in the prediction of halo CME geoeffectiveness and was superior to the DC or PCA model alone.