Shape-based Feature Engineering for Solar Flare Prediction

Solar flares are caused by magnetic eruptions in active regions (ARs) on the surface of sun. These events can have significant impacts human activity, many which be mitigated with enough advance warning from good forecasts. To date, machine learning-based flare-prediction methods employed physics-based attributes AR images as features; more recently, there has been some work that uses features deduced automatically deep learning (such convolutional neural networks). We describe a suite novel shape-based extracted magnetogram Sun using tools computational topology and geometry. evaluate these context multi-layer perceptron (MLP) network compare their performance against traditional attributes. show abstract outperform chosen experts, combination two feature sets improves forecasting capability even further.