GAT · full_fcg
Standard SplitFeb 18, 2026910db720ee184ee1831edeecb931b481
Description
Train GAT on the full_fcg dataset to complete the 3-model x 2-dataset baseline grid.
Conclusion
Matches GIN internal_only as the strongest baseline (98.2% accuracy, 96.9% recall, 99.4% AUROC). GAT benefits more from full_fcg than GIN or GCN. However, all six experiments use standard random splits and may overestimate generalization.
Test Metrics
Accuracy
98.2%
F1 Macro
97.8%
F1 Malware
96.9%
Precision
96.9%
Recall
96.9%
AUROC
99.4%
Best Val Loss
0.1023
Training Time
3645.5000s
Confusion Matrix
| Pred Benign | Pred Malware | |
|---|---|---|
| Actual Benign | 75 | 1 |
| Actual Malware | 1 | 31 |
Configuration
| Hidden Dim | 128 |
| Num Layers | 3 |
| Dropout | 0.5 |
| Batch Size | 4 |
| Learning Rate | 0.001 |
| Weight Decay | 0.0001 |
| Max Epochs | 200 |
| ES Patience | 20 |
| ES Min Epochs | 100 |
| LR Patience | 10 |
| LR Factor | 0.5 |
| Mixed Precision | Yes |
| Random Seed | 42 |
| Epochs Trained | 113 |