Table 6. Optimized hyperparameters of phase 1 of the GCNN model development workflow.
| Hyperparameters | Optimized hyperparameters – phase 1 | ||||
|---|---|---|---|---|---|
| #1 | #2 | #3 | #4 | #5 | |
| Graph convolution layers | [64, 64] | [64, 64, 64] | [128, 64, 32] | [64, 64] | [64, 64, 64] |
| Dense layer size | 256 | 128 | 128 | 64 | 64 |
| Batch size | 16 | 32 | 32 | 32 | 16 |
| Dropout | 9.28 × 10−2 | 6.35 × 10−2 | 7.00 × 10−2 | 1.27 × 10−1 | 3.78 × 10−2 |
| Learning rate | 5.21 × 10−4 | 3.36 × 10−3 | 6.82 × 10−4 | 2.07 × 10−2 | 2.96 × 10−3 |
| Weight decay | 10−3 | 10−3 | 10−4 | 10−5 | 10−5 |
| Activation function | ReLU | ReLU | ReLU | ReLU | ReLU |
| Validation ROC-AUC score | 0.886 | 0.884 | 0.896 | 0.893 | 0.888 |