How to change the classifiers of P300 and SSVEP scenarios?

[willianmrs]

Hello,

I'm starting to work with OpenViBE using EEG-SMT and I am with some difficulties.
My goal is to test the performance (accuracy) of classifiers used by P300 speller, SSVEP and SSVEP Mind Shooter scenarios. Therefore, I want to change the classifier (LDA by default) to SVM and MLP, but I'm getting bad results. So my doubts are:

1. Can I change easily the classifiers of these scenarios? If not, what I need to change to make it work?
2. What parameters can I change to get better results?
3. I'm new with EEG-SMT. So, what are the best channel positions to use with these scenarios?

Thank you in advance.

[nfoy]

Hi,

With the P300 speller, the processing pipeline includes an XDAWN spatial filter to enhance the classification with LDA.
It may have less effect with other classifiers.

When working with P300 ERP, the typical electrodes used are: Fz, Cz, C3, C4, Pz, P3, P4, P7, P8, Oz, O1, O2, but it can also work with other electrodes.

Best regards

[willianmrs]

Hi, thanks for your help.

Actually I'm using the P300 Speller scenario without XDAWN spatial filter because I have only 2 channels, so I'm not using spatial filters.

I managed use the SVM classifier in the P300 Speller scenario but only with linear kernel. When I change the kernel to RBF the classifier doens't work very well, all the answers that I receive contains the same probability, independent of input.

All answers of classifiers processor box has the same probability.

RBF_kernel_error.png (23.31 KiB) Viewed 3500 times

The same happens when I use MLP classifier. Somebody knows whats is wrong?

Best regards.