Improving Speller BCI performance using a cluster-based under-sampling method

A Brain-Computer Interface (BCI) allows its userto interact with a computer or other machines by only usingtheir brain activity. People with motor disabilities are potentialusers of this technology since it could allow them to interact withtheir surroundings without using their peripheral nerves, helpingthem regain their lost autonomy. The P300 Speller is one of themost popular BCI applications. Its performance depends on itsclassifier’s capacity to identify and discriminate the presence ofthe P300 potentials from electroencephalographic (EEG) signals.For the classifier to do this correctly, it is necessary to train itwith a balanced data-set. However, as the P300 is usually elicitedwith an oddball paradigm, only unbalanced distributions can beobtained. This paper applies an under-sampling method based onSelf-Organizing Maps (SOMs) on P300 EEG signals looking toincrease the classifier’s accuracy. Two classifying models, a deepfeedforward network (DFN) and a deep belief network (DBN),are tested with data-sets obtained from healthy subjects and post-stroke victims. We compared the results with our previous worksand observed an increase of 7% in classification accuracy for ourmost critical subject. The DBN achieved a maximum classificationaccuracy of 95.53% and 94.93% for a healthy and post-strokesubject, while the DFN, 96.25% and 93.75%.