Intra-subject Invariant Classification Modeling for Spectral Features in EEG Signals Using Decision Fusion Method

Intra-subject variability of the oscillatory activity in EEG signals limits the personal-adaptability of brain-computer interfaces for neurorehabilitation. The main object of this paper is to construct a fused classification model which is robust to the individual differences in the optimal frequency bands for classifying the spectral features into the dual or single tasks. The proposed decision fusion model results in the higher classification accuracy of 6%, compared to the averaged test accuracy of single classifiers using the best performing band as spectral features. Our study expands the usage of EEG spectral features for neuro-rehabilitation systems without selecting a specific frequency range depending on subject, task or environment.