Sparse learning of band power features with genetic channel selection for effective classification of EEG signals

Padfield, Natasha and Ren, Jinchang and Murray, Paul and Zhao, Huimin (2021) Sparse learning of band power features with genetic channel selection for effective classification of EEG signals. Neurocomputing, 463. pp. 566-579. ISSN 0925-2312 (https://doi.org/10.1016/j.neucom.2021.08.067)

[thumbnail of Padfield-etal-Neurocomputing-2021-Sparse-learning-of-band-power-features-with-genetic-channel-selection]
Preview
 Text. Filename: Padfield_etal_Neurocomputing_2021_Sparse_learning_of_band_power_features_with_genetic_channel_selection.pdf
Accepted Author Manuscript
License: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 logo
Download (1MB)| Preview

Abstract

In this paper, we present a genetic algorithm (GA) based band power feature sparse learning (SL) approach for classification of electroencephalogram (EEG) (GABSLEEG) in motor imagery (MI) based brain-computer interfacing (BCI). The band power in the alpha and beta bands was extracted from the EEG segments and used as features to construct the SL dictionary, in which the GA was employed for channel selection. The GABSLEEG system was tested in three functional areas: i) classification of MI data and idle state data; ii) performance with decreased training data size; and iii) computational efficiency. The system was evaluated by dividing the data into training, validation, and testing sets. The proposed GABSLEEG model is found to significantly outperform conventional classifiers, including the support vector machine (SVM) classifier in (i-iii), and the random forest (RF) and the k-nearest neighbour (k-NN) classifiers in (i-ii). The GABSLEEG system consistently had a higher classification accuracy, sensitivity, and specificity. The average accuracy of the proposed system was 99.65%, on BCI Competition IV dataset 1 and 96.08% for BCI Competition III dataset IVa with the idle state included as a class, which was on a par with state-of-the-art SL and even deep learning approaches.

CORE (COnnecting REpositories)