Feature Selection from Multivariate Time Series Data: A Case Study of Solar Flare Prediction

Abstract

Solar physicists frequently use solar magnetic field parameters for analyzing and predicting solar events. Temporal observation of magnetic field parameters, i.e., multivariate time series (MVTS) representation facilitates finding relationships of magnetic field states to the occurrence of extreme solar events (e.g., solar flares). Feature selection of MVTS-represented solar magnetic field parameters (features) can select the most relevant parameters that give high prediction accuracy. In this paper, we propose a deep learning-based feature selection method, more specifically, an LSTM-based incremental feature selection method, as an end-to-end solution for feature selection in MVTS data. We performed LSTM-based feature selection for multivariate time series data in two steps. Firstly, each MVTS feature is evaluated individually by an LSTM-based univariate sequence classifier, and secondly, the top-performing features are combined to produce input for a downstream LSTM-based multivariate sequence classifier. We compared the proposed MVTS feature selection method with three other baseline feature selection methods on an MVTS-based solar flare prediction dataset and demonstrated that our method selects more discriminatory features compared to other methods.