A comparison of dimensionality reduction techniques for the P300 response

Although P300 is a fairly stable response and therefore utilized in a wide variety of Brain Computer Interface (BCI) systems, the problems of feature selection and dimensionality reduction still constitute a major setback for the applications. In this study, we focus on the selection of best features of P300 data for decreasing the computation time, improving accuracy and visualizing both the underlying classification process and neurophysiological mechanism. To this end, the performance of three feature selection techniques are evaluated. The three techniques are Principle Component Analysis, Spatial Filters for Event Related Potentials and Recursive Channel Elimination. They are applied on the data set acquired through 4-class P300 experiments conducted on 5 subjects. The accuracy profile along with the computational issues are discussed.