Table 1.
Classification accuracy based on different runs of tactile BCI use.
| Data set | Classifier trained on runs # | Classifier tested on runs # | 800 ms post-stimulus | 1000 ms post-stimulus | 1200 ms post-stimulus | 1400 ms post-stimulus |
|---|---|---|---|---|---|---|
| Long stimulus, short ISI | [1 2 3 4] | [5] | 50 | 75 | 75 | 75 |
| [1 2 3 5] | [4] | 25 | 25 | 75 | 100 | |
| [1 2 4 5] | [3] | 50 | 75 | 100 | 75 | |
| [1 3 4 5] | [2] | 100 | 100 | 100 | 100 | |
| [2 3 4 5] | [1] | 50 | 100 | 100 | 100 | |
| Mean ± STD | 55.5 ± 27.4 | 75.0 ± 30.6 | 90.0 ± 13.7 | 90.0 ± 13.7 | ||
| Long stimulus, medium ISI | [1 2 3 4] | [5] | 75 | 100 | 100 | 100 |
| [1 2 3 5] | [4] | 75 | 50 | 100 | 100 | |
| [1 2 4 5] | [3] | 50 | 50 | 75 | 50 | |
| [1 3 4 5] | [2] | 0 | 0 | 50 | 25 | |
| [2 3 4 5] | [1] | 50 | 50 | 75 | 50 | |
| Mean ± STD | 50 ± 30.6 | 50 ± 35.4 | 80 ± 20.9 | 65 ± 33.5 | ||
| Short stimulus, long ISI. Number of stimulations reduced offline for direct comparison | [1 2 3 4] | [5] | 75 | 75 | 50 | 75 |
| [1 2 3 5] | [4] | 75 | 50 | 75 | 100 | |
| [1 2 4 5] | [3] | 75 | 75 | 75 | 75 | |
| [1 3 4 5] | [2] | 25 | 75 | 50 | 0 | |
| [2 3 4 5] | [1] | 75 | 75 | 75 | 75 | |
| Mean ± STD | 65 ± 22.4 | 70 ± 11.2 | 65 ± 13.4 | 65 ± 37.9 | ||
| Short stimulus, long ISI. Full data set with twice as much stimuli. | [1 2 3 4] | [5] | 100 | 75 | 100 | 100 |
| [1 2 3 5] | [4] | 75 | 75 | 75 | 100 | |
| [1 2 4 5] | [3] | 75 | 75 | 75 | 75 | |
| [1 3 4 5] | [2] | 75 | 75 | 75 | 100 | |
| [2 3 4 5] | [1] | 75 | 75 | 50 | 50 | |
| Mean ± STD | 80.0 ± 11.2 | 75.0 ± 0.0 | 75.0 ± 17.7 | 85.0 ± 22.4 | ||
We trained SWLDA classifiers on every combination of four of five runs and tested them on the remaining run. The table furthermore presents classification outcome for classifier weights trained on 800 ms of data, 1000 ms, 1200 ms, and 1400 ms respectively.