Table 2.
Complete offline classifier comparison.
| Ipsi. (I) | Contra. (C) | Bilat. (B) | |||
|---|---|---|---|---|---|
| OVERALL (%) | |||||
| EMG | 8.66 [0.44] | 8.83 [0.41] | (0.51) | 5.01 [0.29]* | (2.57 × 10−7) |
| GONIO | 7.13 [0.79] | 7.69 [0.74] | (0.35) | 2.80 [0.41]* | (2.89 × 10−6) |
| IMU | 3.63 [0.35] | 3.98 [0.31] | (0.09) | 2.26 [0.23]* | (3.43 × 10−5) |
| ALL-LDA | 2.09 [0.27] | 2.48 [0.27] | (7.19 × 10−3) | 1.43 [0.24]* | (2.91 × 10−6) |
| ALL-SVM | 2.33 [0.30] | 2.62 [0.32] | (0.20) | 1.64 [0.26]* | (1.12 × 10−4) |
| ALL-ANN | 3.07 [0.28] | 3.77 [0.29]* | (1.52 × 10−3) | 2.63 [0.26] | (6.26 × 10−3) |
| *α = 0.05/16 | p-value | p-value | |||
| STEADY-STATE (%) | |||||
| EMG | 6.61 [0.41] | 6.76 [0.32] | (0.55) | 3.26 [0.22]* | (1.89 × 10−6) |
| GONIO | 6.59 [0.85] | 7.16 [0.75] | (0.38) | 2.26 [0.38]* | (1.21 × 10−5) |
| IMU | 2.60 [0.31] | 2.92 [0.28] | (0.13) | 1.33 [0.19]* | (3.34 × 10−5) |
| ALL-LDA | 1.25 [0.19] | 1.38 [0.17] | (0.24) | 0.76 [0.14]* | (6.27 × 10−5) |
| ALL-SVM | 1.00 [0.17] | 1.14 [0.16] | (0.30) | 0.71 [0.14] | (8.95 × 10−3) |
| ALL-ANN | 1.71 [0.16] | 2.04 [0.21] | (0.02) | 1.55 [0.14] | (0.15) |
| *α = 0.05/12 | p-value | p-value | |||
| TRANSITIONAL (%) | |||||
| EMG | 17.83 [0.78] | 18.18 [1.22] | (0.67) | 12.91[0.85]* | (3.02 × 10−7) |
| GONIO | 9.51 [0.82] | 10.06 [0.94] | (0.46) | 5.18 [0.63]* | (2.79 × 10−6) |
| IMU | 8.31 [0.74] | 8.70 [0.64] | (0.36) | 6.40 [0.56]* | (6.08 × 10−4) |
| ALL-LDA | 5.94 [0.84] | 7.42 [0.82]* | (9.70 × 10−4) | 4.50 [0.76]* | (5.19 × 10−4) |
| ALL-SVM | 8.37 [1.04] | 9.30 [1.22] | (0.19) | 5.84 [0.89]* | (6.18 × 10−5) |
| ALL-ANN | 9.16 [0.93] | 11.52 [0.97]* | (6.94 × 10−4) | 7.46 [1.00]* | (3.91 × 10−3) |
| *α = 0.05/12 | p-value | p-value | |||
Error rates (mean [SEM]) are shown for all evaluated classifiers. Asterisks under contralateral denote statistically significant differences between contralateral and ipsilateral sensor sets (p-value). Asterisks under bilateral denote statistically significant differences between bilateral and ipsilateral sensor sets (p-value). The best-performing classifier for each type of error is bolded.