Rotational data augmentation for electroencephalographic data.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Deep learning on image data commonly uses data augmentation techniques like scaling and elastic deformations.
• Electroencephalographic (EEG) data, unlike image data, faces significant challenges due to insufficient training datasets.

Purpose of the Study:

• To propose and evaluate rotational distortions as a novel data augmentation method for EEG signals.
• To improve the performance of EEG-based brain-computer interfaces through enhanced data augmentation.

Main Methods:

• Applying rotational distortions, analogous to image augmentation techniques, to generate artificial EEG data.
• Specifically, rotating EEG data around the y- and z-axes by approximately ±18 degrees.

Main Results:

• The proposed rotational distortion method significantly increases the performance of signal processing chains for EEG-based brain-computer interfaces.
• Generated data through rotation around specific axes proved to be meaningful and effective for augmentation.

Conclusions:

• Rotational distortions offer an efficient and effective approach for EEG data augmentation.
• This work encourages the exploration of further novel methods for augmenting EEG data to improve machine learning model performance.