Classifiers Fusion for EEG Signals Processing in Human-Computer Interface Systems

In this paper we study the effectiveness of using multiple classifier combination for EEG signals classification aiming to obtain more accurate results than it possible from single classifier system. The developed system employs different features vectors fused at the abstract and measurement levels for integrating information to reach a collective decision. For making decision, the majority voting scheme has been used. While at the measurement level, fuzzy integral, majority vote, decision template and some other types of combination methods have been investigated. The ensemble classification task is completed by feeding the Support Vectors Machines with Redial Basis Kernel functions classifiers with different features extracted from the EEG signal for imagination of right and left hands movements (i.e., at EEG channels C3 and C4). The parameters of SVM classifiers were optimized by genetic algorithm. The results show that using classifier fusion methods improved the overall classification performance.