Common Spatial Patterns-Low classification results,Help needed

I have two EEG datasets of 14*5120*60(numChnumSamplesnumTrials)-one ERD and another neutral(both classes are filtered between 8-13Hz). I used common spatial patterns algorithm to get the projection matrix W and extracted first 3 and last 3 columns of W (denoted it as W0-to make 14*6 matrix). I projected each trial(x=14*512) onto W0, W0'*x and got 6*512. I calculated the variance of each row, so that it gave me a 6_d vector. Like wise I got 120(60 for ERD and 60 for Neutral) 6-D vectors. All the neutral trials are assigned to class 0(when neutral trial is projected onto W,6-D vector is assigned to class 0) and ERd trails are assigned to class 1(when ERD trial is projected onto W,6-D vector is assigned to class 0). I used 10-Fold cross validation SVM, but ironically I have got only 45-55% classification accuracy. I don't know whee I have made an error, Can anyone one help me out? Many thx in advance.