Randomized trials in EEG classification experiments are mandatory as drift is pervasive

Hari Bharadwaj, Ronnie Wilbur, Jeffrey Mark Siskind

Keywords: EEG

TL;DR: We demonstrate that about 100 papers are flawed because they don't use randomized trials

Abstract:

Temporal correlation is demonstrated in three public EEG datasets.
Filtering does not remove the correlation. The community is cautioned
that a substantial number of recent publications are flawed due to a
confound between temporal correlation and stimulus class, despite
claims that filtering removes the correlation. The only known way to
avoid this confound is through randomization.

Primary Area: applications to neuroscience & cognitive science

Submission Number: 4793