Evolutionary perspective for optimal selection of EEG electrodes and features

This paper proposes a novel evolutionary approach to the optimal selection of electroencephalogram (EEG) electrodes as well as relevant features for effective classification of cognitive tasks. The EEG electrode and feature selection (EFS) problem here has been formulated in the framework of an optimization problem with an aim to simultaneously satisfying four criteria. The first criterion deals with maximization of the correlation between the selected features of EEG source signals, before and after the selection of optimum electrodes. It thus ensures the preservation of information of the cortical sources corresponding to a cognitive task even after reducing the number of electrodes. The second criterion is concerned with minimization of the mutual information between the selected features of the EEG signals recorded by the selected electrodes. It helps in identifying the unique information by reducing the redundancy in the EEG signals recorded by the selected electrodes for a specific cognitive task. The third criterion aims at optimal selection of EEG electrodes and EEG features in an attempt to i) minimize the difference between the selected EEG source-features (to ensure their similarity) for a specific cognitive task and ii) maximize the difference between the selected EEG source-features (to ensure the efficient categorization) of different cognitive tasks. The last criterion is concerned with maximization of the classification accuracy of different cognitive tasks based on the selected EEG source-features, corresponding to the selected EEG electrodes. The originality of the paper lies in obtaining the sets of optimum EEG electrodes and EEG features by independent optimization of individual objectives. These sets of optimum EEG electrodes and EEG features are then ranked based on their fuzzy memberships to satisfy individual four objectives. A self-adaptive variant of firefly algorithm (referred to as SAFA) is proposed to optimize individual objectives by proficiently balancing the trade-off between the computational accuracy and the run-time complexity. Experiments undertaken over wide variety of cognitive tasks reveal that the proposed algorithm outperforms the other standard algorithms (applied to the same problem) in terms of accuracy and computational overhead. (C) 2017 Elsevier Ltd. All rights reserved.