Contributor: Cyrus Shahabi
Co-contributor(s): Kiyoung Yang, Hyunjin Yoon
Research Lab:  Infomation Laboratory, University of Southern California

In order to perform the channel selection, we applied Corona proposed in [1]
on the row differences, not on the raw data. Each of EEG signals was 
transformed into a vector using the upper triangle elements of its 
correlation coefficient matrix. Both the max and mean aggregating functions 
have been utilized. As a classifier, Support Vector Machine (SVM) with 
linear kernel has been employed both for the channel selection and the 
cross validation.

For each subject, the training and testing have been separately performed, not
utilizing the data from other subjects.
 
Subject FSS_method          # of selected channels  CV accuracy
aa      Corona (mean, corr)      32                  81.4881%
al      Corona (mean, corr)      61                  96.1607%
av      Corona (mean, corr)      18                  77.9518%
aw      Corona (mean, corr)      33                  83.2143%
ay      Corona (mean, corr)      42                  80.0000%

[1] Kiyoung Yang, Hyunjin Yoon, Cyrus Shahabi, A Supervised Feature Subset 
    Selection Technique for Multivariate Time Series, International Workshop 
    on Feature Selection for Data Mining: Interfacing Machine Learning with 
    Statistics (FSDM) in conjunction with 2005 SIAM International Conference 
    on Data Mining (SDM'05), Newport Beach, CA, April 2005