Table 3. Performance comparison using m-sequence stimulation for cVEP BCI.
| Subject | EEG2Code | OCSVM [22] | |
|---|---|---|---|
| trained on random sequences | trained on m-sequences | trained on m-sequences | |
| S1 | 100.0% | 100.0% | 100.0% |
| S2 | 94.8% | 97.9% | 98.9% |
| S3 | 50.0% | 39.0% | 53.7% |
| S4 | 100.0% | 99.5% | 100.0% |
| S5 | 98.4% | 98.4% | 98.4% |
| S6 | 95.3% | 96.3% | 96.8% |
| S7 | 97.4% | 98.4% | 97.9% |
| S8 | 86.5% | 91.6% | 93.2% |
| S9 | 84.9% | 95.8% | 99.5% |
| mean | 89.7% | 90.8% | 93.2% |
Shown are the accuracies (ACC) of the target prediction on data where m-sequences were used for stimulation. For the EEG2Code we used the model trained on random stimulation patterns during the online experiment on the one hand and the models trained on m-sequence trials using a randomized 10-fold cross-validation (CV) on the other hand. The results are compared to the results using the one-class support vector machine (OCSVM) also by using a randomized 10-fold CV. The results are averaged over 192 trials with a duration of 1.05 s.