| Abstract : | Solar eruptions such as CMEs, flares disrupt geomagnetic and communication systems on Earth. While flares are abrupt, bright events that occur in the solar atmosphere and emit massive amounts of energy in the 10^28 to 10^32 erg range, CMEs are intense eruptions that hurl plasma into interplanetary space. CMEs can be found in conjunction with flares, filaments, or independent. Although both flares and CMEs are understood as triggered by a common physical process magnetic reconnection, yet, the degree of association is unknown. In this work, we study the time series of magnetogram data derived from SHARP (Space weather HMI Active Region Patches) to understand eruptive flare mechanism using Machine Learning models SVM,LDA and Deep learning model LSTM. Here, we use 18 SHARP parameters as input to our Machine Learning model from the year 2011-2021. The task here is to perform binary classification, hence two classes, predicting whether a flare will be accompanied by CMEs or not. We initially attempt to study the features at different time lags that will be more responsible for eruptive flare. For example, MEANSHR shows deviated mean between two classes at 48h time lag, while MEANGBZ shows it at 8-24h time lag before the event occurence. Therefore, we determine the appropriate time lag for each feature using our Deep Learning model LSTM, coupled with ML models SVM and LDA, to perform binary classification. We further attempt to study the model’s predictions and behaviour using Explainable ML methods such as variable-importance measure and shapley. |
