README.md

metadata

language:
  - az
  - en
license: mit
tags:
  - cybersecurity
  - phishing-detection
  - url-classification
  - threat-intelligence
metrics:
  - precision
  - recall
  - roc_auc
  - pr_auc
  - mcc
  - brier_score
model-index:
  - name: urlaz
    results:
      - task:
          type: url-classification
          name: Phishing URL Detection
        metrics:
          - type: precision
            value: 0.998
          - type: recall
            value: 0.955
          - type: pr_auc
            value: 0.9983
          - type: roc_auc
            value: 0.9989

🛡️ URLAZ — Phishing URL Detection Engine

URLAZ is a lightweight, high-performance Machine Learning model built for real-time URL-based phishing detection.

---

📊 Benchmark Metrics (5-Fold GroupKFold Cross-Validation)

Metric	Score	Description
Precision	99.80%	Test set precision at operational threshold
Recall	95.50%	Phishing detection recall
PR-AUC	0.9983	Precision-Recall Area Under Curve
ROC-AUC	0.9989	Receiver Operating Characteristic AUC
MCC	0.9781	Matthews Correlation Coefficient
Brier Score	0.0078	Probability Calibration Score

---

💻 Quickstart (Python)

1. Download Model from HuggingFace

from huggingface_hub import hf_hub_download
import joblib

# Download model weights
model_path = hf_hub_download(repo_id="alixansec/urlaz", filename="urlaz_phishing_detector.joblib")
model = joblib.load(model_path)

2. Predict URL

# Pass 35 structural features extracted from URL (see predict_url.py)
probability = model.predict_proba([features])[0][1]

if probability >= 0.95:
    print("🔴 PHISHING DETECTED")
else:
    print("🟢 SAFE")

---

📁 Repository Contents

• urlaz_phishing_detector.joblib — Serialized binary classifier
• urlaz_phishing_detector.sha256 — SHA-256 integrity signature
• phishing_urls_verified.txt — Verified targeted phishing dataset (3,067 records)
• predict_url.py — Inference prediction script

---

🔒 Integrity Signature

• SHA-256: c6a21e5a06901d6f3ba848a2d6c8507ff48c2cb52b2d8d80da6717a142c0a445