Evolutionary Multiobjective Feature Selection in Multiresolution Analysis for BCI

Abstract

Although multiresolution analysis (MRA) may not be considered as the best approach for brain-computer interface (BCI) applications despite its useful properties for signal analysis in the temporal and spectral domains, some previous studies have shown that MRA based frameworks for BCI can provide very good performance. Moreover, there is much room for improving the performance of the MRA based BCI by feature selection or feature dimensionality reduction. This paper investigates feature selection in the MRA-based frameworks for BCI, proposes and evaluates several wrapper approaches to evolutionary multiobjective feature selection. In comparison with the baseline MRA approach used in previous studies, the proposed evolutionary multiobjective feature selection procedures provide similar or better classification performance, with significant reduction in the number of features that need to be computed.