Brain-computer interface: Frequency domain approach using the linear and the quadratic discriminant analysis

Brain-computer interface (BCI) offers solutions for those with severe neuromuscular disorder. Indeed, it provides new non-muscular channel to control external devices. The aim of this work is to increase the classification accuracy rate using the suitable system parameters and methods. In this present study Welch's method for power spectral density (PSD) estimation has been used for features extraction followed by two different classification methods (Linear Discriminant Analysis (LDA) and Quadratic Discriminant Analysis (QDA)). The task was to think about right and left hand movement. A study of the influence of the flowing parameters was performed: frequency bands, predictive features, and classification method. The results show that the most significant increase takes place by improving the PSD estimation. Selecting the specific frequency bands of each cortical area provides also an important improvement. Finally the use of the suitable classifier is essential to attain optimal performances.