about the accuracy

Forum for discussion on different user applications
Yueqing Li
Posts: 51
Joined: 08 May 2007, 16:41

about the accuracy

Post by Yueqing Li » 11 Oct 2007, 13:30

Hi, Dr. Schalk:

We are doing P300 Speller experiments these days. Every time, the accuracy is below 20%, sometimes even less than 5%. However, the graph produced by Mario is really good for some subjects. The impedance is below 10k for the first 4 channel, while much higher for the last 4 channel (we use 8 channel according to the document). However, the wrong spelled characters are really very close to the target (general they are close to each other) for these subjects.

For some subjects, the graph is very weird. In the r square plot, there is even no red (blue and green everywhere), so did the topograph. However, for some subjects, it is all yellow or red in the r square plot; in the topograph, it is all deep red. For these subjects, the wrong spelled characters are always far away from the target.

All the above graph is produced by Mario downloading all the data files of one subject (9 sessions, 4 characters each session).

I want to know what is a normal accuracy range for P300 Speller experiment. Also, can I use the best performance channel (it will change in every session, but often Cz) to represent the result? I found the word predicted by the best performance channel is with a higher accuracy rate than the accuracy we got from 8 channels on the whole.



Who is online

Users browsing this forum: No registered users and 2 guests