Skip to Main Content
A recently proposed boosting algorithm is Boosting Neural Networks, which improves the performance of single neural classifiers. In this paper, A powerful methods is introduced to automatically detect P300 subcomponents in multi-channel electroencephalogram (EEG) trials. Firstly, EEG data are projected to a subspace based on independent component analysis where the P300 is more obvious, and then P300-related independent components are selected according to the a priori knowledge of P300 temporal pattern, and clear P300 peak is reconstructed by back projection of ICA. Secondly, Boosting Neural Networks is employed in the detection of P300 subcomponent. This detection is based on the temporal and spatial patterns in the EEG trials after ICA-Based subspace enhancing. These methods can be used to build a brain-computer interface, such as a Farwell and Donchin's P300-based speller paradigm. The algorithm described here is tested off-line with dataset II from the BCI Competition 2005. Our results indicate that the Boosting Neural Networks has an effectiveness and efficiency performance for both subjects, and show that the temporal features of P300 is more stable and discriminable than spatial features.