International Conference on Computer and Knowledge Engineering

Home / 15th International Conference on Computer and Knowledge Engineering

Enhancing EEG-based BCI Performances by Reducing Covariate Shift via Adaptive Multi-Domain Feature Extraction

Authors :

Moein Radman¹ Reza Arghand² Nader Nariman-Zadeh³ Ali Chaibakhsh⁴

1- University of Essex 2- University of Guilan 3- University of Guilan 4- University of Guilan

Keywords :

Brain-computer interface،Adaptive feature extraction،Covariate shift minimization،Constant-Q FBCSP،Probabilistic Classification Vector Machines

Abstract :

The main goal of this paper is to improve the functional accuracy of brain-computer interface (BCI) systems by addressing the challenges created by non-stationary EEG signals in certain subjects. To deal with this problem, the EEG signals are decomposed into several frequency bands using a bank of Constant-Q filters, as the best features from the temporal, spectral, and spatial domains are extracted. These features are used to train the probabilistic classification vector machines, where covariate shift minimization is used to adapt the features. The assessment phase is based on the BCI 2008-2b competition dataset, which results in achieving a Kappa score of 0.78 and performance enhancement of about 8.7 and 12.2 percent for subject one and subject two, respectively.