Single-Trial Decoding of Scalp EEG under Natural Conditions

Figure 4

Test accuracies for a classifier trained on pseudotrials (averaged categories) (black) or trained on single-trials (red) and tested on single-trials. The x-axis refers to the hold-out test subject. Chance level prediction accuracy is 0.565 (dashed line). Note. In some cases, the “optimum parameters” are not found to be optimum, which can be explained by different training phases of the two single-trial classifiers. The classifier based on validation sets was trained on 13 subjects while the classifier with parameters based on the test set was trained on 14 subjects. For 5 out of 15 subjects, the classifier based on 13 subjects was able to obtain higher accuracies.