Table 5 Comparative performance of hybrid ML–PINN framework versus classical RSM on external test set.
Model | Target | R2 | RMSE | MAE |
|---|---|---|---|---|
Gradient Boosting Regressor (GBR) | LE% | 0.89 | 6.24 | 4.87 |
Physics-Informed Neural Network (PINN) | LE% | 0.85 | 8.10 | 6.25 |
Response Surface Methodology (CCD) | LE% | 0.74 | 10.80 | 8.50 |
Gradient Boosting Regressor (GBR) | EE% | 0.87 | 7.15 | 5.42 |
Physics-Informed Neural Network (PINN) | EE% | 0.83 | 8.10 | 6.25 |
Response Surface Methodology (CCD) | EE% | 0.71 | 12.30 | 9.80 |
