Spatial Pattern of Electroencephalography (EEG) Extracted by Nonlinear Features during Working Memory Maintenance

When working on electroencephalography (EEG) of cognitive tasks, there is a problem of high channel dimension. In this work, a channel selection method is proposed to achieve efficient spatial pattern. First, it extracts the linear and nonlinear multi-features of the entire brain EEG signal and applies the Support Vector Machine (SVM), K-Nearest neighbors (KNN), Random Forest (RF) multiple classifiers for classification. It is found that the nonlinear feature has a better classification performance in the memory material classification task (images vs characters) based on scalp EEG signals of the working memory maintenance procedure. Then the SVM is applied for classification on each channel by using the nonlinear feature. The top-10 channels with better classification performance are proved as the specific spatial pattern of the working memory maintenance procedure based on working memory materials. The channel selection method proposed in this paper can be extended to other classification analysis of cognitive tasks to extract the efficient spatial patterns.