VEPS - Vulnerability Exploitation Prediction Score

A machine learning pipeline for predicting the likelihood of CVE exploitation using NVD data, CVE observations, and multiple models including DistilBERT and XGBoost.

Overview

VEPS processes National Vulnerability Database (NVD) data to predict which CVEs are most likely to be exploited in the wild. The system combines:

• Data Processing: Automated NVD data download and feature extraction
• DistilBERT Models: For predicting missing CVSS vectors and CWE classifications
• XGBoost Model: For vulnerability exploitation likelihood prediction
• Daily Pipeline: Automated daily predictions on new CVE data

# Download latest NVD data
uv run src/scripts/data/download_data.py --latest

# Extract features from NVD data
uv run src/scripts/veps/extract_features.py

# Create training dataset
uv run src/scripts/veps/create_training_set.py

# Perform hyperparameter search (optional)
uv run src/scripts/veps/hyperparameter_tuning.py

# Train vulnerability prediction model
uv run src/scripts/veps/train_pipeline.py

# Make daily predictions
uv run src/scripts/veps/prediction.py

Required input files

CVE Observations(data/cve_observations.csv)

cve,date,count
CVE-2021-44228,2021-12-10,15
CVE-2021-44228,2021-12-11,42

CVE mentions (data/cve_mentions.json) (optional)

{
  "CVE-2021-44228": {
    "2021-12-09": 5,
    "2021-12-10": 23
  }
}

Output Format

Daily predictions saved as predictions_YYYYMMDD.csv

cve_id,base_score,attack_vector,...,exploitation_probability,percentile
CVE-2024-12345,9.8,NETWORK,...,0.94,99.2

Cyberrisk

VEPS was developed in the Cyberrisk project.

The research project “Cyberrisk” is a joint partnership between mnemonic, Norsk Regnesentral, Avinor and DNB.

The Research Council of Norway has granted the research project as part of their Innovation Project for the Industrial Sector, a funding instrument that provides grants to business-led innovation projects that make extensive use of research and development (R&D). The Innovation Project is to lead to renewal and sustainable value creation for the project’s business partners, and funding should also generate socioeconomic benefits by making new knowledge and solutions available.