Table 5.
Scoring function and root mean squared error (RMSE) before and after using dynamic weighting (DW) for loss function while maintaining the architecture of deep learning models. Blue colored text indicates improved performance while red colored text indicates worsened performance. The values in this table are the median values across 20 experimental runs.
| Deep Learning Architecture |
Scoring Function | RMSE |
|---|---|---|
| Bidirectional LSTM | 178.568 | 20.1 |
| Bidirectional LSTM + DW | 129.089 | 13.9 |
| −27.7% | −30.6% | |
| DNN | 93,473.3 | 23.1 |
| DNN + DW | 13,741.3 | 23.9 |
| −85.2% | +3.4% | |
| CNN1D | 112.858 | 22.3 |
| CNN1D + DW | 63.002 | 21.1 |
| −44.1% | −5.7% | |
| Bidirectional GRU | 169.550 | 11.6 |
| Bidirectional GRU + DW | 81.899 | 12.9 |
| −51.6% | +11.8% |