EEG + EOG artifact (eye blink) removal + RNN

[MasumeTaherii]

Hi,

I am working on a project focused on eye-blink artifact removal from EEG signals using a recurrent neural network (RNN). I am using the dataset provided in this repository.

However, I am facing an issue where the model does not seem to learn effectively. The output signal for EEG segments containing eye-blink artifacts is almost identical to the clean EEG signal, which suggests that the model is not successfully removing the artifacts.

I would like to ask:

• Are there any recommended preprocessing steps for this dataset (e.g., normalization, windowing, or filtering)?
• Is there a baseline RNN architecture or training setup that works well for eye-blink artifact removal?
• Are there specific loss functions or evaluation metrics you recommend for this task?
• Is this dataset suitable for supervised artifact removal, or are there any known limitations?

Any advice or guidance would be greatly appreciated.
Thank you for sharing this dataset.