Risk Fusion MLP — Standard
Component 10a | Final Year Project
Fuses GNN + CVE + identity scores into risk score (0-10) with feature attention gate showing which input drove the decision. Hard category cutoffs. Trained on 2000 real NVD CVEs.
Quick Start
from risk_fusion import RiskFusionPipeline
pipe = RiskFusionPipeline.from_hub("sohomn/risk-fusion-mlp-standard")
result = pipe.predict(structural=0.78, temporal=0.82, cvss=9.8,
exploitability=3.9, impact=5.9, identity_anomaly=0.65)
# Returns Feature Weights showing which input mattered most
Performance
| Metric | Value |
|---|---|
| Test MAE | 0.0836 |
| Test RMSE | 0.1057 |
| Test R2 | 0.9924 |
License
MIT - Research use only.