Abstract
There are a lot of researches on Brain-Computer Interface which could be used for robot control and text input by analyzing users' electroencephalogram. In general, classifier(s) is used for EEG classification such as Linear Discriminant Analysis(LDA), and it needs to be trained with some known data before classification. For improving the classification accuracy, a large number of training data is necessary, and measuring it often gives a lot of burdens to the user. Thus, the purpose of this research is to reduce the number of data required for training.
This paper proposes a retraining approach which based on classified data along with training data, and investigates its performance on BCI text input system called P300 speller. The P300 speller provides "Backspace" which could delete the misclassified letters, thus it could be assumed that and inputted letter has been correctly classified if it has not been deleted by the Backspace. In the proposed approach, such letters could be used for retraining. The proposed approach could make it possible to reduce the number of training data.
