A Hierarchical Classification Model for Solar Flare Prediction

Solar flares can have significant damaging effects on Earth's space environment, space weather, technological and biological systems. In this paper, a hierarchical classification model is proposed for predicting solar flare from time series of solar magnetic field parameters, which builds multiple gradient tree boosting models trained with different feature sets in the first stage and improves the performance of these base models through supervised decision fusion using a Naive Bayes classifier in the second stage. The proposed method has been utilized for flare prediction in BigData Cup Challenge 2019, which has achieved the 5^th scores in the preliminary stage and a weighted rank of 8^th in the finalists.