IRIS Multiple Instance Learning Dataset

This dataset contains the data for the paper 'Using Multiple Instance Learning for Explainable Solar Flare Prediction'.

It consists of 10'000 spectrographs (bags) recorded by NASA's IRIS satellite with associated class labels AR (non-flaring active region) and PF (pre-flare active region). Each spectrograph consists of several hundred individual spectral pixels (instances).

The dataset is intended to explore the use of multiple instance learning for the prediction of solar flares. Even though class labels are only known at the level of the full spectrograph ('bag-level'), multiple instance learning allows to learn the association of individual spectra to each of the classes ('instance-level').

More information is provided in the paper and on the Zenodo page.