Classification of motor imagery tasks for BCI with multiresolution analysis and multiobjective feature selection.

Julio Ortega,John Q Gan,Andrés Ortiz,Javier Asensio-Cubero

doi:10.1186/s12938-016-0178-x

Abstract

BackgroundBrain-computer interfacing (BCI) applications based on the classification of electroencephalographic (EEG) signals require solving high-dimensional pattern classification problems with such a relatively small number of training patterns that curse of dimensionality problems usually arise. Multiresolution analysis (MRA) has useful properties for signal analysis in both temporal and spectral analysis, and has been broadly used in the BCI field. However, MRA usually increases the dimensionality of the input data. Therefore, some approaches to feature selection or feature dimensionality reduction should be considered for improving the performance of the MRA based BCI.MethodsThis paper investigates feature selection in the MRA-based frameworks for BCI. Several wrapper approaches to evolutionary multiobjective feature selection are proposed with different structures of classifiers. They are evaluated by comparing with baseline methods using sparse representation of features or without feature selection.Results and conclusionThe statistical analysis, by applying the Kolmogorov-Smirnoff and Kruskal–Wallis tests to the means of the Kappa values evaluated by using the test patterns in each approach, has demonstrated some advantages of the proposed approaches. In comparison with the baseline MRA approach used in previous studies, the proposed evolutionary multiobjective feature selection approaches provide similar or even better classification performances, with significant reduction in the number of features that need to be computed.

Highlights

Brain-computer interfacing (BCI) applications based on the classifica‐ tion of electroencephalographic (EEG) signals require solving high-dimensional pattern classification problems with such a relatively small number of training patterns that curse of dimensionality problems usually arise
Multiobjective optimization in supervised feature selection Our multiobjective optimization procedure for feature selection has been implemented through a wrapper approach, which can be seen as a search for the feature set that optimizes a cost function that evaluates the utility of the given features according to the performance attained by the classifier
This section presents the experimental results obtained by the evolutionary multiobjective feature selection approaches, in comparison with the baseline methods described in the previous section

Summary

Introduction

Brain-computer interfacing (BCI) applications based on the classifica‐ tion of electroencephalographic (EEG) signals require solving high-dimensional pattern classification problems with such a relatively small number of training patterns that curse of dimensionality problems usually arise. Brain computer interfaces (BCI) try to identify the cognitive states of the user to control a computer or any other kind of devices These systems promote interesting and useful applications based on this new way of human–machine communication, such as those related with the improvement of quality of life for people with disabilities. The curse of dimensionality is usually present in the classification of EEGs as the number of patterns (EEGs) available for training is relatively small, and the number of features is usually much larger than the number of available training patterns This way, as many other high-dimensional pattern classification or modeling tasks, BCI requires feature selection techniques in order to remove redundant, noise-dominated, or irrelevant inputs. Feature extraction may be more efficient in dimensionality reduction, but it usually results in a loss of the interpretability of the new feature space

Methods

Results

Conclusion