We propose an approach to electroencephalogram feature selection and classification problems in brain-computer interfaces based on a committee of weak classifiers. The design of a classification committee is formulated as an optimization problem and the greedy algorithm for its solving is considered. The proposed approach is applicable when the objects to be classified are characterized by a large number of features while a few train samples are available. Classification performance of the committee was evaluated on real data and improvement over traditional classification methods was observed.
展开▼
