ABSTRACT
The recent advancements in natural language processing
(NLP) have introduced novel artificial intelligence models for
data classification, extending their scope to analyzing brain
signals acquired via electroencephalogram (EEG). Among
these developments, the transformer architecture, which has
become available in recent years, has provided researchers
with a powerful model to explore and evaluate its capabilities
in various EEG-related studies, including developing
new assistive devices tailored for individuals with impaired
communication skills. This work leverages the transformer
model to classify P300 event-related potentials on publicly
available EEG data, aiming to benchmark its accuracy against
established algorithms documented in literature. Upon conducting
the case study, the results reveal that the transformer
achieves a noteworthy accuracy rate of 95%, indicating its
viability as a classifier for P300-based spellers.