Research Article
A New Approach to Noninvasive-Prolonged Fatigue Identification Based on Surface EMG Time-Frequency and Wavelet Features
Table 5
Classification results of prolonged fatigue based on the naïve Bayes method.
| Parameter | Muscles | Features | Performance | BF | RF | VL | VM |
| Time features (ΔMAV, ΔRMS) | Accuracy (%) | 70 | 78 | 64 | 56 | Specificity (%) | 100 | 84 | 83 | 97 | Precision (%) | 0 | 67 | 36 | 0 | CVErr | 0.31 | 0.25 | 0.43 | 0.44 |
| Frequency features (ΔFmed, ΔFmean) | Accuracy (%) | 86 | 95 | 68 | 77 | Specificity (%) | 88 | 94 | 89 | 83 | Precision (%) | 79 | 96 | 36 | 69 | CVErr | 0.15 | 0.04 | 0.39 | 0.23 |
| Time and frequency features (ΔMAV, ΔRMS, ΔFmed, ΔFmean) | Accuracy (%) | 94 | 98 | 95 | 97 | Specificity (%) | 97 | 100 | 100 | 97 | Precision (%) | 86 | 96 | 88 | 96 | CVErr | 0.06 | 0.01 | 0.07 | 0.02 |
| Wavelet index features (ΔWIRM1551, ΔWIRM1M51, ΔWIRM1522, ΔWIRE51, ΔWIRW51) | Accuracy (%) | 82 | 91 | 80 | 66 | Specificity (%) | 85 | 93 | 78 | 71 | Precision (%) | 77 | 89 | 84 | 58 | CVErr | 0.18 | 0.09 | 0.23 | 0.38 |
| Time, frequency, and wavelet index features (ΔMAV, ΔRMS, ΔFmed, ΔFmean, ΔWIRM1551, ΔWIRM1M51, ΔWIRM1522, ΔWIRE51, ΔWIRW51) | Accuracy | 87 | 88 | 77 | 90 | Specificity | 86 | 91 | 79 | 83 | Precision | 89 | 85 | 80 | 100 | CVErr | 0.16 | 0.15 | 0.23 | 0.2 |
|
|