Training study approaches for a SVM-based BCI: Adaptation to the model vs adaptation to the user

Abstract

Support Vector Machine (SVM) is extensively used in BCI classification. In this paper this classifier is used to differentiate between two mental tasks related to motor imaginary in order to check the possibility of improvement with two alternative adaptation (user's adaptation and model adaptation). Two kind of training have been done by 4 subjects. In the first test (user's adaptation to the model), each subject use a personalized model and 7 sessions are registered to compare the evolution of the results due to the user's training. This initial model is done with a preliminary session which include register of 6 different motor imaginary tasks to select the best combination of them. The second test (model's adaptation to the user) tries to evaluate the benefits of the updating of the model with new registers. The results show that, at least for this kind of imaginary tasks, these methods of adaptation with a SVM-based system have not a meaningful increase of the success rate. © 2013 Springer-Verlag.