BCI competition 2005: DatasetV

Group information:

Institute of Biomedical Engineering of Xi’an Jiaotong University

Xi’an, 710049, P. R. China

Team Member: Xiaomei Pei, Guangyu Bin


Supervisor: Prof. Chongxun Zheng.


1.     Description of our method

The precomputed features are used to classify the three mental tasks.

Firstly, the 8 consecutive PSD samples in each frame of 62.5ms are averaged t in every 0.5s.

Secondly, according to the Fisher ratio of the averaged PSD within the three classes, the most reactive PSD features are selected to form the new feature vectors for further classification.

Finally, by the kernel-based Fisher Discriminant Analysis, the three mental tasks could be discriminated. The kernel parameters are selected by cross validation to minimize the classification error. 

2. Format of the data

The result is saved in a Matlab-file format. sub1.mat, sub2.mat, sub3.mat are the classifier output with labels 2, 3, 7 of the test data for subject1, 2, 3 respectively.