Berita BPI LIPI
oleh : , Keum-Shik Hong
Abstract. In this paper, a new adaptive neural network classifier (ANNC) of EEGP300 signals from mental activities is proposed. To overcome an overtraining of the classifier caused by noisy and non-stationary data, the EEG signals are filtered and their autoregressive (AR) properties are extracted using an AR model before being passed to the ANNC. For evaluation purposes, the same data in Hoffmann et al. (2008) were used. With and without the AR property extraction, the proposed ANNC could achieve 100% accuracy for all the subjects. To verify the performance improvement of the proposed classification scheme, a comparison of the ANNC and the conventional back-propagation neural network classifier was performed as well.
Keywords: Adaptive high order neural network, EEG-P300 potentials, Feature extraction, Classification, Brain computer interface