Table 9 Performance comparison of TL-LSTM variants to evaluate the effects of Multi-Head Attention (MHA) and feature selection.
Model | Architecture | Feature selection | MAE | RMSE | R2 | Purpose |
|---|---|---|---|---|---|---|
A | TL-LSTM-MHA | Yes | 4.38 | 5.80 | 0.9974 | Full proposed model |
B | TL-LSTM (No-MHA) | Yes | 12.77 | 18.23 | 0.9742 | Ablation: no attention |
C | TL-LSTM—MHA | No | 5.16 | 6.94 | 0.9963 | Ablation: no FS |
