Hello, I have a large EEG dataset for which I need to do Independent Component Analysis (ICA)? Any leads on how to do it?

I have got EEG dataset (98765 X 25 matrix) for which I need to do ICA.

Answer 1

[Raunak Gupta]

Hi,

For Reducing the dataset exact Independent Component Analysis (ICA) doesn’t exist but you may use rica for doing the dimensionality reduction. The Reconstruction ICA work closely as ICA but an added feature of optimizing non-linear objective function that also include penalty while reconstructing the output. This can improve the convergence speed as compared to ICA and the features you get are globally orthogonal. Reconstruction ICA is supported in R2017a and later versions.

For more information about Reconstruction ICA Algorithm you may look Feature Extraction Algorithm.

For Other Dimensionality reduction methods you may look Dimensionality Reduction and Feature Extraction.

Answer 2

[Mandavi Gahlot]

Thank you so much for your answer, but I need to keep the dimensions same as reducing its dimensions will affect the labelled data. I wish to first do the ICA and then extract features for specific range of values.

[Raunak Gupta]

Hi,

From the EEG Dataset, I can understand that each data-point contains 25 features. Out of that I am assuming that last feature is label so effectively there will be 24 features so by doing rica mentioned above the number of feature for each data point can be bring down to less than 24. That way the information about the labels will not lost since each datapoint will have less than 24 features and one label. You may try to seperate labels first and then apply rica on 98765 x 24 matrix so the dimensionality can be reduced.

Hope this helps.