Towards User-Friendly Spelling with an Auditory Brain-Computer Interface: The CharStreamer Paradigm
Figure 6
Classification accuracy for the calibration data of three conditions.
The binary classification accuracy, estimated with cross validation is plotted for each condition and subject (A). The thick black line marks the mean. Plots B depicts the multi-class accuracy for the two classification approaches (“std” and “meta”). This was estimated by cross-validation on calibration data, using entire trials as test sets. Precisely, the point for rank = i quantifies the fraction of trials with a rank of the target class equal or lower than i. Thus, the mean multi-class performance (correct decision – rank = 1) was 47% (41%) for the meta (std) classifier. One can observe that 77% (72%) of the trials have a multi-class rank better or equal than 5. While perfect BCI control (each 30-class decision is correct) would result in a straight line with y = 1, the dashed line marks the multi-class accuracy based on chance level.