From: Multiclass EEG motor-imagery classification with sub-band common spatial patterns
Author | No. of channel | Dataset | Feature extraction | Feature selection | Results | ||
---|---|---|---|---|---|---|---|
Data type | Classifier | Accuracy | |||||
Nicolas Alonso et al. (2015) [16] | 22 | Dataset 2a, BCI Competition IV | CSP | MIBIF | Binary class | SRLDA | 85% |
Multi class | 74% | ||||||
Shiratori et al. (2015) [17] | 15 | Dataset acquired by themselves from 8 healthy subjects | CSP | Mutual information | Finger tapping | Random forest (RF) | 88.7 ± 4.5% |
Motor imagery | 56.7 ± 4.4% | ||||||
Yong and Menon, (2015) [19] | 32 | Dataset acquired by themselves from 12 healthy persons | CSP, FBCSP, and band power | – | Binary class | SVM | 80.5% |
Multi class | 60.7% | ||||||
Shiman et al. (2017) [20] | 32 | Dataset acquired by themselves from 9 healthy persons | FBCSP | – | 3 classes | LDA | 69.1 ± 7.9% |
4 classes | 62.8 ± 6.8 % | ||||||
Ge, Wang and Yu, (2014) [28] | 60 | Dataset IIIa open BCI competition | STFT and CSP | – | FP2 channel | SVM | 78.3% |
C4 channel | 88.1% | ||||||
She et al. (2015) [22] | 22 | Dataset 2a, BCI competition IV | CSP | – | Multiclass | SVM | 48.4% |
NBPW | 53.8% | ||||||
NBPW along FBCSP | 59.3% | ||||||
PPTSVM | 62.4% | ||||||
Gao et al. (2016) [23] | 64 | Dataset acquired by themselves from 10 healthy subjects | Kolmogorov complexity | – | Multiclass | ELM | 73.0% |
Adaboost ELM | 79.5% | ||||||
Meisheri et al. (2018) [25] | 22 | Dataset 2a, BCI competition IV | CSP | – | Multiclass Data | SRT2NFIS using JAD | 74.65% |
Proposed algorithm for multiclass classification by SBCSP-SBFS | 8 and 14 | Wet gel rlectrodes and Emotiv Epoc | SBCSP | SBFS | Emotiv Epoc | NBPW | 60.61% |
Wet gel electrodes | KNN | 86.50% |