COMPARING COMMON MACHINE LEARNING CLASSIFIERS IN LOW-DIMENSIONAL FEATURE VECTORS FOR BRAIN COMPUTER INTERFACE APPLICATIONS

There are lots of classication and feature extraction algorithms in theeld of brain computer interface. It is signicant to use optimal classication algorithm and fewer features to implement a fast and accurate brain computer interface system. In this paper, we evaluate the performances ofve classical classiers in different aspects including classication accuracy, sensitivity, specicity, Kappa and computational time in low-dimensional feature vectors extracted from EEG signals. For our experiments we used BCI Competition 2003 Data Set III and Data Set Ia. Classiers were compared on 61 different datasets which were created with a combination of extracted features. When classiers were ranked based on the average values of performance metrics, we conclude that the NB and the SVM classiers are shown to be good candidates for pattern classi�- cation for low dimensional feature vectors. On the other hand, it can obviously mention that decision tree classier provides the worst performance. We believe that this paper has a signicant contribution in theeld of classier for brain computer interface appli- cations.