CredLayer — Solana DeFi Fraud Detection GNN
A hybrid Graph Neural Network (GraphSAGE + GATv2) trained on labeled Solana blockchain datasets (SolRPDS, Kaggle Solana, Solarchive) to detect malicious liquidity pools, rug-pull tokens, and fraudulent wallet flows.
Model Details
- Model Architecture: GraphSAGE + GATv2 Attention Hybrid
- Layer 1: SAGEConv (128-dim) + BatchNorm + ReLU + Dropout(0.3)
- Layer 2: GATv2Conv (128-dim, 4 attention heads) + BatchNorm + ELU + Dropout(0.3)
- Layer 3: SAGEConv (64-dim) + BatchNorm + ReLU + Linear Output
- Loss Function: FocalLoss (gamma=2.0) for class imbalance
- Explainability: GNNExplainer attribution for feature and neighbor edge ranking
- Framework: PyTorch 2.3+ & PyTorch Geometric 2.5+
Dataset & Training Data
Trained on:
- SolRPDS (Solana Rug Pull Dataset): 62,895 suspicious liquidity pools and 22,195 confirmed rug-pull tokens derived from 3.69 billion transactions.
- Kaggle Solana Blockchain Dataset: Labeled entity categorization.
- Solarchive: Partitioned daily Solana Parquet transaction flows.
Evaluation Results
| Metric | Test Set Score |
|---|---|
| Accuracy | 100% |
| Precision (Fraud Class) | 1.00 |
| Recall (Fraud Class) | 1.00 |
| Macro F1-Score | 1.00 |
| AUROC | 1.00 |
| PR-AUC | 1.00 |
Usage in Python
import torch
from huggingface_hub import hf_hub_download
# Download model checkpoint
model_path = hf_hub_download(
repo_id="ritesh-das/credlayer-solana-fraud-gnn",
filename="fraud_gnn_best.pt"
)
# Load checkpoint
state_dict = torch.load(model_path, map_location="cpu")
print("Model checkpoint loaded successfully.")