Comparison of Principal Component Analysis and Partial Least Square Discriminant Analysis in the Classification of EEG signals

摘要

Brain Computer Interface (BCI) is the scientific advent to use human brain signals to control computerized systems or other external devices. Here, we propose a signal processing-based approach for the classification of Electroencephalogram (EEG) signals acquired from the human brain during the movement of a feedback bar to the left and right directions. The dataset used to this work is from the BCI competition II. Our proposed model applies two multivariate regression algorithms known as Partial Least Square (PLS) and Principal Component Analysis (PCA) coupled with Discriminant Analysis (DA) for the classification of the subject feedback session. Lowpass band filters along with baseline correction and smoothing techniques such as asymmetric least squares and Savitzky-Golay transformation are used to preprocess the EEG signals before classification. Results indicate that PCA-DA as a classifier outperforms PLS-DA with an accuracy of 82.14%.