Hi everyone,
I performed classification as tutorial suggested, I used two different conditions with 112 trials from single subject and chose "classifier with permutation", it ended up without error. But when I opened the result which turned out some negative decoding accuracy that really confused me a lot.
Indeed, a negative accuracy is not possible. Looking at your graph, it feels to me like you have applied an online filter that changes the actual accuracy value. If you go to the Filter tab in Brainstorm, can you make sure that you are not applying a high-pass filter? You can use a low-pass to smooth out your line a bit but using a high-pass would drastically change your range of values.
If that doesn't help, can you confirm that the actual values of the matrix are negative? Right click on your decoding matrix and go to File -> Export to Matlab, and then look at the "Value" vector.
Yes, it was the "Filter" issue, after turning off the "high-pass filter" option, everything looks well now and I was excited about the robust decoding result.
Indeed, we might be interested in applying a low-pass filter to these decoding accuracy values, but never a high-pass filter.
I changed the logic of the Filter tab panel, hoping to reduce the risks of wrong interpretations:
I removed the mirroring option (outdated with the current filtering functions)
I renamed the last option into "Filter all results", and this box has now to be explicitly checked to get the decoding results (and all the other "matrix" files) filtered by online filters.