Table 2 Parameters of proposed MOBO approach during training and testing.
Parameters | Ranges explored | Range considered for training | Range considered for testing and validation | Final selection criteria |
|---|---|---|---|---|
Learning rate | [0.0001, 0.001, 0.005] | 0.0001 | 0.0001 | Lowest RMSE on validation |
Batch size | [32, 64, 128] | 64 | 64 | Best convergence behaviour on validation |
GNN Layers | [1, 2, 3] | 2 | 2 | Best accuracy & RMSE on validation |
Hidden units (EAGRU & GNN) | [32, 64, 128] | 64 | 64 | Trade-off between accuracy and computation time |
Dropout rate | [0.1, 0.3, 0.5] | 0.3 | 0.3 | Overfitting occurred with the value 0.1 hence the next value is chosen |
Attention heads | [1, 2, 4] | 4 | 4 | Best temporal feature learning occurs with this attention head |
Graph sampling ratio | [0.3, 0.5, 0.7, 1.0] | 0.5 | 0.5 | Best spatial representation value is achieved over there |
Optimization epochs | 100 (with early stopping) | 95 | 95 | RMSE stagnation on validation set |
