After analyzing the single-channel, single-stimulus type
classifiers, we sought to determine whether classifier accuracy
could be improved by allowing multiple channels and/or
stimulus types to vote on identification. This analysis was
based on the fact that, by design, the different stimulus types
likely tap into different functional brain networks, each one
contributing unique variability to a user’s full set of ERP
responses. Figure 5 illustrates this concept, by displaying scalp
maps of classification accuracy for the black and white food,
celebrity, and oddball color targets classifiers. As would be
expected, maximum classification accuracy for the oddball
color targets is found a broad area originating on the back of
the head, which is in accordance to the distribution of the P300
component. The black and white food, in contrast, are best
classified over the most occipital part of the head. Celebrities
are best classified over a group of channels intermediate
between the oddball and food areas of best classification.
Thus, the regions of scalp that provide strong single-stimulus
type classification accuracy differ across stimulus types, which
suggests that combining information from multiple stimulus
types and / or channels might improve classifier accuracy.