Solar Flare Prediction Model with Three Machine-learning Algorithms using Ultraviolet Brightening and Vector Magnetograms

Abstract

We developed a flare prediction model using machine learning, which is optimized to predict the maximum class of flares occurring in the following 24 hr. Machine learning is used to devise algorithms that can learn from and make decisions on a huge amount of data. We used solar observation data during the period 2010-2015, such as vector magnetograms, ultraviolet (UV) emission, and soft X-ray emission taken by the Solar Dynamics Observatory and the Geostationary Operational Environmental Satellite. We detected active regions (ARs) from the full-disk magnetogram, from which ∼60 features were extracted with their time differentials, including magnetic neutral lines, the current helicity, the UV brightening, and the flare history. After standardizing the feature database, we fully shuffled and randomly separated it into two for training and testing. To investigate which algorithm is best for flare prediction, we compared three machine-learning algorithms: the support vector machine, k-nearest neighbors (k-NN), and extremely randomized trees. The prediction score, the true skill statistic, was higher than 0.9 with a fully shuffled data set, which is higher than that for human forecasts. It was found that k-NN has the highest performance among the three algorithms. The ranking of the feature importance showed that previous flare activity is most effective, followed by the length of magnetic neutral lines, the unsigned magnetic flux, the area of UV brightening, and the time differentials of features over 24 hr, all of which are strongly correlated with the flux emergence dynamics in an AR.

Predicting Solar Flares Using SDO/HMI Vector Magnetic Data Products and the Random Forest Algorithm

“…We also compare our results with previous works on forecasting flares larger than M1.0 class. By looking at the TSS score, it is apparent that the performance of our method coupling SDO/HMI magnetic parameters with RF (TSS ≈ 0.53) is very similar to Bloomfield et al (2012) and Ahmed et al (2013), while being ∼28% lower than Bobra & Couvidat (2015) and Nishizuka et al (2017). Nonetheless, we obtain quite high scores in both recall and precision.…”

“…Nonetheless, we obtain quite high scores in both recall and precision. We note that Nishizuka et al (2017) used the ERT classifier similar to RF, and also considered, besides magnetic field parameters, UV brightenings and previous flare activity to achieve high metric scores.…”

“…Many previous studies (e.g., Ahmed et al 2013;Bobra & Couvidat 2015;Nishizuka et al 2017) include non-flaring ARs as negative samples. This difference in building data sets with multiple classes may affect the performance comparisons.…”

SeMiner: A Flexible Sequence Miner Method to Forecast Solar Time Series

“…This is the reason why those methods do not consider the evolution of a solar data time series in the mapping, which causes the loss of valuable information for the forecasting process. Some forecasting methods set solar data observed in a specific instant of time to the class of a solar flare occurred after the observed data [7][8][9]18,19]. However, there are also works that map subseries of the solar data into events observed in the future, so that they adequately consider the historical evolution of solar data [10,11].…”

“…Considering classes C, M, and X, some methods consider "Positive" forecasts for classes greater than or equal to "C" [7], others consider "Positive" for forecasts greater than or equal to a class M solar flare [8][9][10][11], and others forecast individual probabilities for each class (C, M, X).…”