An EEG-based Brain-Computer Interface for Attention State Recognition

When people are in different attention states, electroencephalogram(EEG) signals will be different. Most of researchers used the power of brain waves in different frequency bands or their power ratio as features for classification. However, the accuracy of recognition needs further improvement. In this paper, we proposed a BCI system used for attention state classification. Firstly, we used a band-pass filter to filtered the raw EEG data. After preprocessing, eight candidate features were extracted based on the power spectral density (PSD). Then we used a genetic algorithm(GA) based on probability distribution to select eight-dimensional features from the twenty-four-dimensional candidate features. Finally, we chose support vector machine (SVM) as classifier. The average accuracy of the ten subjects achieved 86.99% for the frontal(the central area and parietal area achieved 84.94% and 84.66%, respectively). Further analysis, the classifier was changed to a back-propagation neural network (BP). Output of the bp classifier was classified using the k-means. The final average accuracy of the ten subjects achieved 90.30% for the frontal(the central area and parietal area achieved 89.66% and 88.99%, respectively). We also used the 24-dimensional PSD feature to calculate the accuracy, the proposed method can greatly improve the accuracy, which showed the feasibility of this method in attentional EEG analysis.