Fig. 2

Graphical illustration of the three cross validation methods. The ‘Leave-one-out’ method use all epochs from eight subjects for training, and all epochs from the remaining subject for testing. This is done for all nine subjects (ninefold cross validation). In the ‘Total’ method, all epochs from all subjects are pooled together and partitioned into 20 subsets. The classifier is trained on 19 subsets and tested on one subset. This is done for all 20 combinations of subsets (20-fold cross validation). In the ‘Individual’ method the classifier is subject specific. For each subject, the epochs are partitioned into ten subsets, the classifier is trained on nine subsets and tested on one subset. Thereby the algorithm is validated ten times on each of the nine subjects (90-fold cross validation)