Table 11 Performance comparison of the models with different number of channels.
From: A multi-scale feature extraction fusion model for human activity recognition
Models | WISDM | UCI-HAR | PAMAP2 | ||||||
|---|---|---|---|---|---|---|---|---|---|
Accuracy (%) | F1-score (%) | Parameters | Accuracy (%) | F1-score (%) | Parameters | Accuracy (%) | F1-score (%) | Parameters | |
One-channel model | 93.34 | 93.53 | 208,719 | 94.74 | 94.68 | 209,121 | 94.20 | 94.18 | 212,647 |
Two-channel model | 95.23 | 95.19 | 416,286 | 95.86 | 95.87 | 417,102 | 94.78 | 94.74 | 423,595 |
Three-channel model (proposed) | 97.18 | 97.17 | 623,987 | 96.71 | 96.72 | 625,229 | 96.28 | 96.27 | 634,775 |
Four-channel model | 95.39 | 95.39 | 831,822 | 95.22 | 95.23 | 833,502 | 95.97 | 95.96 | 846,187 |
