Weighted extreme learning machine for P300 detection with application to brain computer interface

Brain–computer interface (BCI) is a communication system, which brain signals can be analyzed and transformed into commands of external devices. The P300-speller BCI utilizes the P300 evoked potentials, which generated in oddball paradigms, to spell characters. In order to eliminate the imbalance existed between target and non-target stimuli data in electroencephalogram (EEG) signals, we apply weighted extreme learning machine (WELM), which consequently improves the P300 detection accuracy. For starter, the raw EEG signals of all channels are divided into several segments by a given time window. Afterwards, features of each segment are extracted by principal component analysis. Then, WELM executes a classification for above features. Last, outputs of WELM classifier are used to recognize the target character in the P300 speller. Experiments were carried out based on the third BCI Competition Data Set II and achieved an average accuracy of 97%, and it still reached 85% with the first eight high sensitivity channels for P300 responses.