CICIDS GAT Diagnoser (Stage 2)
Graph Attention Network for multiclass intrusion detection. Classifies network attacks into 8 unified threat families using graph topology.
Performance
- Macro F1: 90.60%
- Weighted F1: 96.15%
- Classes: 8 Threat Families
Architecture
- Type: Heterogeneous Graph Attention Network
- Node Embeddings: 64-dimensional
- Hidden Layers: 128-dimensional
- Attention Heads: 4
- Edge Features: 58 flow statistics
- Regularization: Dropout(0.3)
Usage
from utils.model_loader import HybridDiagnoserGAT, load_gat_model
model = load_gat_model()
pred_class, confidence, probs = predict_gat(model, features, src_ip, dst_ip)
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support