Authors:
Alexey Syskov
;
Vasilii Borisov
;
Vsevolod Tetervak
and
Vladimir Kublanov
Affiliation:
Ural Federal University named after the first President of Russia B.N. Yeltsin, Russian Federation
Keyword(s):
Accelerometer, Brain-Computer Interface, Electroencephalography, Machine Learning, Mental Evaluation, Test of Variables of Attention, Principal Component Analysis.
Abstract:
In the paper the results of extracting and selection the features of EEG data and accelerometer for mental status evaluation are shown. We have used 14 channel wireless EEG-system Emotiv EPOC+ with accelerometer (motional data - MD) for short-term recording under several functional states for 10 healthy subjects: Functional rest (rest state), TOVA-test (mental load), Hyperventilation (physical load) and Aftereffect (after test state). We then extracted core features from EEG-only and MD-only data using principal component analysis. After that, supervised learning methods were used for mental state classification: EEG-only core features for AF3, T7, O1, T8, AF4 channels, MD-only core features and EEG- MD integrated core features. Experimental results showed that integrated core features for mental status evaluation have higher prediction accuracy 92,0% for decision tree method.