COMPARISON OF FEATURE STAGES IN A MULTI-CLASSIFIER BCI
Authors:
Ivan J. CESTER
Neurochemistry Group, Physiological Science I, University of Barcelona, Spain
Aureli SORIA-FRISCH
Starlab Barcelona SL, Spain ivan.cester@ub.edu | aureli.soria-frisch@starlab.es
Abstract:
The present communication is focused on the comparison of two different approaches for the realization of the feature extraction stage in a Brain-Computer Interface (BCI) system. For this purpose we compare the usage of Common Spatial Patterns (CSP) and of Wavelet Analysis (WA). Due to the large dimensionality of the wavelet feature space, we further compare the performance of two selection procedures: Analysis of Variance (ANOVA) and Genetic Algorithms (GA). The inclusion of such data-driven feature selection stages avoids the application of a priori hypothesis on the most relevant frequency bands. An extensive performance analysis is given for supporting the outperformance of the WA approach over CSP, and of the GA one over ANOVA.
Link: