The document summarizes publicly available MI-EEG datasets released between 2002 and 2020, sorted from newest to oldest. This document also summarizes the reported classification accuracy and kappa values for public MI datasets using deep learning-based approaches, as well as the training and evaluation methodologies used to arrive at the reported results.
The content of this document is part of the following review paper:
Altaheri, H., Muhammad, G., Alsulaiman, M. et al. Deep learning techniques for classification of electroencephalogram (EEG) motor imagery (MI) signals: a review. Neural Comput & Applic 35, 14681–14722 (2023). https://doi.org/10.1007/s00521-021-06352-5
The following figure shows the performance of EEG-based motor imagery (MI) classification reported by the latest deep learning-based articles for all public MI datasets.
HO: Hold-out (train: test), CV: Cross-validation, LOSO: Leave-one-subject-out, c-sub: Cross-subject, sub-d: Subject-dependent, sub-i: Subject-independent, sb: subjects, “(x : y sb)”: x subjects for training and y subjects for testing
For the details, the reader can refer to the above article.