---
license: apache-2.0
tags:
  - graph-neural-network
  - rgcn
  - edge-attention
  - binary-classification
---

# BFS_EA_RGCN Model

This is a Graph Neural Network model combining Residual Graph Convolutional Networks (RGCN) and Edge Attention mechanisms, designed for binary classification of graphs (vulnerable vs. non-vulnerable).

- **Architecture**: Classifier with res_GCN (9 layers, 11 hidden units) and Edge_Attention (2 heads, 100 feature length)
- **Task**: Binary classification of software graphs
- **Dataset**: SG_Final_Train and SG_Final_Test (vulnerable and non-vulnerable graphs)
- **Input**: Graph node features (100-dim), edge features (100-dim), adjacency matrix
- **Output**: Binary label (0: non-vulnerable, 1: vulnerable)
- **Checkpoint**: Loaded from /kaggle/input/egat-peculiar-model/checkpoint_epoch_4.pt
- **Performance**: 
  - Test Accuracy: ~0.5959
  - Precision: ~0.6345
  - Recall: ~0.4623
  - F1 Score: ~0.5349
  - AUC: ~0.6222 (from exp-0 in notebook)

For more details, refer to the original training notebook or dataset documentation.