VulnPriority Risk Ranker
LightGBM LambdaRank model for CVE prioritization refinement within policy tiers.
Intended Use
- Refine vulnerability ordering within the same policy tier
- Complement (not replace) transparent policy-based ranking
Limitations
- Trained on synthetic/offline CVE corpus snapshot
- Not a substitute for live NVD API enrichment in production
- Does NOT use CVSS×EPSS as a scoring method
Training
- Temporal split: train < 2022-06, val < 2023-06, test ≥ 2023-06
- Objective: LambdaRank (NDCG)
- Features: cvss, epss, epss_percentile, kev_status, patch_available, asset_exposure, asset_criticality
By Aria AI Security Engineering Team.