Table 7 Performance comparison with alternative ML architectures.
From: Intelligent optimization of track and field teaching using machine learning and wearable sensors
Architecture | F1-score | MAE | Training time | Parameters |
|---|---|---|---|---|
Proposed CNN-BiLSTM | 0.94 ± 0.01 | 0.083 | 4.2 h | 2.3 M |
Transformer-based | 0.89 ± 0.02 | 0.107 | 8.7 h | 5.1 M |
Standard LSTM | 0.86 ± 0.02 | 0.124 | 3.1 h | 1.8 M |
Multimodal fusion | 0.91 ± 0.01 | 0.095 | 6.3 h | 3.7 M |
Random forest only | 0.82 ± 0.03 | 0.156 | 1.8 h | 0.9 M |
3D-CNN | 0.88 ± 0.02 | 0.112 | 5.4 h | 4.2 M |
