Table 3 Comparison of proposed method with previous studies using same EEG dataset
Sl. no | Feature selection method | Average classification accuracy | No. of selected channels | Avg. comp time in s | |||
|---|---|---|---|---|---|---|---|
Existing methods | NN | FA | NN | FAM | NN | FAM | |
1 | Spectral ratios of δ to γ band (7 spectral ratios) + GA + NN + FAM classifiers [22] | 94.3 | 81.8 | 7 | 7 | (Train + 200 test vectors classification time only) | |
0.3 | 0.17 | ||||||
2 | γ sub-band power + PCA + k-NN classifier [23] | NN | Not discussed | ||||
95.83 | 61 | ||||||
94.06 | 16 | ||||||
86.01 | 8 | ||||||
75.13 | 4 | ||||||
3 | Mean γ power and correlation coefficient measure between channels + SVM classifier [24] | 80 | 45 | Not discussed | |||
4 | Nonlinear feature extraction (Hurst, Lyapunov exponent, higher-order spectra, ApEn, SaEn) + SVM classifier [25] | 91.7 | 7 | Not discussed | |||
5 | Spectral entropy features + SEPCOR + k-NN + MLP classifier [21] | Correlation threshold | Classification accuracy k-NN | Classification accuracy MLP | SEPCOR feature vectors | Computation time (s) k-NN | Computation time (s) MLP |
0.1 | 99.60 | 93.43 | 22 | 7.30 | 28.55 | ||
95.45 | 30.74 | ||||||
97.55 | 32.55 | ||||||
99.60 | 55.70 | ||||||
0.08 | 99.30 | 89.26 | 15 | 5.22 | 28.31 | ||
91.11 | 30.07 | ||||||
93.35 | 30.04 | ||||||
95.60 | 53.60 | ||||||
6 | Proposed method spectral entropy features with t test ranking +PCA + k-NN + MLP classifier | No of pc. = 25 | |||||
No rank | k-NN | 25 | k-NN | ||||
91.54 | 5.90 | ||||||
Rank | 93.87 | 5.90 | |||||
No of pc. = 15 | |||||||
No rank | 91.96 | 15 | 5.67 | ||||
Rank | 93.08 | 5.67 | |||||