Feature selection for cursor task

[Ryan]

Greeting, guys,

After the feature selection session, I got really bad results. the r-squared value in the high frequency band is really large. I know if there are big r squared value in low frequency band, it's due to the noise; but what is the reason for high frequency band to have large r-squared value?

and also, for the calibration feature selection part, following the cue,the subbject should imagine the movement or perform the real movement, or it doesnot matter?

I check the spectra plot on one channel, and found out the frequency for the biggest r-squared value is not consistent with the frequency that shows the biggest energy drop. I wonder if there is something I did wrong, or there is a proper reason for this phenomenon.

Thanks a lot!
Ran

[jawilson]

Ryan (or Ran?)
It is hard to tell what is going wrong without seeing a plot. What is the "high frequency band?" Is this 100+ Hz? What is is large r^2 value? Also, are you doing a P300 or a Mu-rhythm experiment?