Adaptive dimensionality reduction method using graph-based spectral decomposition for motor imagery-based brain-computer interfaces

In this work, we consider the problem of electroencephalography (EEG) signal classification for motor imagery brain-computer interfaces. The goal is to identify the pattern of the brain activity using a robust method for pre-processing, processing, and classification of the EEG signals. To this end, a new graph-based framework is proposed to reduce the dimensionality of the data by taking into account not only the geometrical structure of the channels/electrodes, but also the correlation between the EEG signals. The most significant feature vectors required for EEG signals classification are adaptively selected through spectral decomposition of the data using the graph Laplacian matrix. The tangent space mapping method is then applied to bring the captured data into Euclidean space. In order to classify the dimensionally-reduced EEG signals, the linear support vector machine algorithm is employed. Experiments are conducted on five different subjects consisting of right hand and right foot motor imagery actions. The results show that the proposed method can provide higher classification accuracy as compared to the other existing methods that we tested.