Skip to main content
. 2026 Sep 11;20:1885665. doi: 10.3389/fnhum.2026.1885665

Table 2.

Classification accuracy (%) of the evaluated models under 10–90% synthetic fatigue using within-subject and strict LOSO protocols.

Method 10% 20% 30% 40% 50% 60% 70% 80% 90%
Within-subject nested stratified five-fold cross-validation
FRFNet 89.9 ± 3.1 88.7 ± 3.4 82.3 ± 4.0 78.1 ± 4.5 73.0 ± 4.8 67.8 ± 5.5 64.3 ± 6.1 62.1 ± 6.8 60.0 ± 7.6
E2FNet 87.2 ± 3.5 85.8 ± 3.8 76.8 ± 4.6 70.6 ± 5.0 63.5 ± 5.3 57.5 ± 6.0 53.2 ± 6.8 50.4 ± 7.5 47.3 ± 8.1
DCA Fusion 84.9 ± 3.8 82.0 ± 4.1 74.2 ± 4.8 67.4 ± 5.2 60.0 ± 5.7 53.7 ± 6.4 50.5 ± 7.2 47.0 ± 8.1 44.8 ± 9.4
DMEFNet-adapted 83.2 ± 3.9 80.4 ± 4.0 75.2 ± 4.6 69.3 ± 5.1 63.5 ± 5.4 57.1 ± 6.2 54.0 ± 7.0 51.2 ± 7.8 48.3 ± 9.0
FAConformer-style 77.2 ± 8.1 71.8 ± 8.4 63.9 ± 7.9 58.6 ± 7.7 54.6 ± 7.8 52.0 ± 8.3 50.8 ± 8.0 49.5 ± 8.2 48.5 ± 7.9
DBF 62.4 ± 9.3 59.3 ± 9.8 54.6 ± 9.9 50.6 ± 9.1 48.2 ± 8.4 45.6 ± 7.1 43.7 ± 6.5 41.7 ± 6.6 40.8 ± 6.4
EEG-only 70.2 ± 4.2 70.2 ± 4.2 70.2 ± 4.2 70.2 ± 4.2 70.2 ± 4.2 70.2 ± 4.2 70.2 ± 4.2 70.2 ± 4.2 70.2 ± 4.2
EMG-only 89.3 ± 3.9 87.7 ± 4.4 80.4 ± 4.8 68.9 ± 5.7 57.0 ± 6.6 48.5 ± 7.3 44.0 ± 8.1 41.3 ± 8.8 39.8 ± 9.8
Strict leave-one-subject-out (LOSO)
FRFNet 71.6 ± 10.3 71.1 ± 10.0 70.0 ± 10.0 68.1 ± 10.6 65.8 ± 10.4 62.7 ± 9.6 60.2 ± 9.7 58.1 ± 10.5 57.0 ± 10.1
E2FNet 61.1 ± 10.7 56.4 ± 11.6 50.6 ± 12.0 45.1 ± 9.7 40.6 ± 6.6 37.5 ± 4.3 35.7 ± 2.9 35.0 ± 2.6 34.8 ± 3.5
DCA Fusion 61.0 ± 11.5 60.5 ± 11.2 59.0 ± 10.8 56.5 ± 10.5 53.2 ± 9.6 49.8 ± 9.0 46.5 ± 8.5 44.0 ± 8.0 42.8 ± 7.8
DMEFNet-adapted 62.4 ± 12.8 61.6 ± 12.8 60.2 ± 12.1 58.2 ± 10.7 54.7 ± 9.8 50.9 ± 9.2 47.4 ± 8.8 45.3 ± 7.9 44.2 ± 7.6
FAConformer-style 59.7 ± 9.8 59.3 ± 9.8 58.5 ± 10.2 57.2 ± 10.1 56.3 ± 9.5 54.8 ± 9.4 53.9 ± 9.1 53.6 ± 9.3 53.1 ± 8.8
DBF 60.1 ± 9.6 58.4 ± 9.1 56.2 ± 8.5 54.8 ± 7.8 53.7 ± 7.3 52.0 ± 7.6 51.1 ± 8.0 50.6 ± 8.5 50.3 ± 8.2

Values are mean ± sample standard deviation across 25 participants. Bold values indicate the proposed FRFNet model, provided as the reference for comparison with each baseline method. EEG-only and EMG-only were evaluated only under the within-subject protocol.