A CSP\AM-BA-SVM Approach for Motor Imagery BCI System

• Published in: IEEE access Vol. 6; pp. 49192 - 49208
• Main Authors: Selim, Sahar, Tantawi, Manal Mohsen, Shedeed, Howida A., Badr, Amr
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.01.2018; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Algorithms; Attractor metagene algorithm; Band-pass filters; Bat algorithm; brain computer interface; Cognitive tasks; Competition; Datasets; Electroencephalography; Feature extraction; Human-computer interface; Image classification; motor imagery; Optimization; Support vector machines; Synchronous motors; Task analysis; Training
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2018.2868178

Brain-computer interface (BCI) has become extremely popular in recent decades. It gained its significance from the intention of helping paralyzed people communicate with the external environment. One of the major challenges facing BCI systems is obtaining reliable classification accuracy of motor imagery (MI) mental tasks. In this paper, a novel CSP\AM-BA-SVM approach is proposed using bio-inspired algorithms for feature selection and classifier optimization to improve classification accuracy of the MI-BCI systems. The proposed approach applies optimum selection of time interval for each subject. The features are extracted from EEG signal using the common spatial pattern (CSP). Binary CSP is extended to multi-class problems by utilizing one-vs-one strategy. This paper introduces applying a hybrid attractor metagene (AM) algorithm along with the Bat optimization algorithm (BA) to select the most discriminant CSP features and optimize SVM parameters. The efficacy of the proposed approach was examined using three data sets. The proposed approach has achieved 78.55% accuracy and 0.71 mean kappa for BCI Competition IV data set 2a, 86.6% accuracy and 0.82 mean kappa for BCI Competition III data set IIIa, and 85% for the binary class BCI Competition III data set IVa. For multi-class data sets, the proposed approach outperforms winners of BCIC IV, 2a and BCIC III, IIIa with kappa 0.14 and 0.17, respectively. For binary class BCIC III, IVa, it performed slightly better than existing studies in the literature by ≈ 0.5%. The proposed CSP\AM-BA-SVM transcends the traditional CSP\SVM approach and other existing studies.