Comparison of EEG Data Classification between Conventional Visual Cue-Marker and EMG-Based Marker on Brain Activity

摘要

In Brain-Computer Interfaces (BCI), data sets need consistency factor to train good filters and classifiers. If there is lack of pattern in the data sets, it will affect the classification accuracy of the results. Consistency in extracted data-epoch requires precise markers to be embedded along the continuous Electroencephalography (EEG) data collection process. This study suggests placing the markers on actual movement through Electromyography (EMG) data classifications during isometric finger flexion and extension. An EMG-based marker system is applied to generate marker in EEG signal during data measurement in real time. EEGLAB software is used for pre-processing. The data loaded into the BCILAB interface for chosen feature extraction and classification. Spectrally weighted Common Spatial Pattern (Spec-CSP) used as feature extraction method and Linear Discriminant Analysis (LDA) is implemented for classification process. Results show that EMG-based marker approach shows higher potential success rate, at 73.6% compared to visual cue-based marker at 71.1%. It is suggested that EMG-based marker approach is applicable in finding pattern recognition of EEG data in isometric finger flexion and extension.