An Effective Classification Approach for EEG-based BCI System

One way to enhance performance of a BCI system is to improve accuracy of classifier. In this paper we apply two development Adaboost classifiers on the basis of an advanced boosting learning algorithm: AdaboostNN and Gentle Adaboost. AdaboostNN works by training nearest-neighbour weak learner on the resampled weighted training data in each iteration, then the weak hypotheses is linearly combined as the final prediction, while a decision tree classifier is available as the weak learner adopted by Gentle Adaboost. LDA and SVM classification methods are also tested to make a comparison with AdaboostNN and Gentle Adaboost. Besides, influence of the number of CSP filters on classification result is also discussed in this paper. By comparison, we get a conclusion that both of these two classifiers are considered to perform more effectively than LDA and SVM, even when the EEG features get a lower separability between two classes.