Feature Optimization in ERP-Based Brain Computer Interface Utilizing Adaptive Boosting

In current studies on ERP-based brain computer interface, a big challenge is to find subjectspecified feature combination for more robust classification. In this paper we propose a recursive feature optimization method based on adaptive boosting and support vector machine to select optimal feature combination. The results of ERP-based brain computer spelling experiment on 11 subjects prove that AdaBoost-based optimization method can significantly improve classification accuracy and simultaneously depress feature dimension greatly. Meanwhile the computational complexity of optimization method is simplified by AdaBoost strategy for practical possibility.