Solar Flare Forecast Magnetogram Dataset

For a given forecasting time window (12, 24, or 48 hours), magnetograms within the specified time window before a flare event are treated as positive samples, while negative samples are randomly selected from magnetograms without flare events. The final dataset spans from May 2010 to January 2019 and contains 2,988 positive samples and 70,822 negative samples. A nonoverlapping temporal splitting strategy is adopted, using data from 2011 and 2015 as the testing set and the remaining years as the training set, in order to avoid data leakage and ensure reliable model evaluation.

The dataset is highly imbalanced, with positive-to-negative sample ratios of approximately 1:25 in the training set and 1:22 in the testing set.