A bagging SVMs to learn from Event Related Potentials using Electroencephalography

摘要

In this paper, we present an efficient approach to investigate data of EEG-based Brain-Machine Interface (BMI) using a bagging Support Vector Machines (SVMs) for collected data classification from a P3-speller paradigm. The combination of SVMs allows to handle the problem of EEG data variability between the different sessions of the acquisition process. This variability is caused by temporal non-stationarity of brain EEG signals. Hence, each SVM classifier is applied to a training set from the same session BMI data acquisition. A temporal extraction of P3 Event Related Potentials features is firstly conducted, then an offline classification of the extracted P3-speller responses is performed using the mixed SVMs. Simulation results show good performances classification, the ratio reach 84% after 15 test sequences.