PhishDetector

End-to-end phishing URL detection project with:

• A Flask ML API for classifying URLs.
• A static web UI for quick checks.
• Data prep + training scripts for model updates.

Live Demo

https://aiphishdetector.netlify.app/

What’s Inside

• backend/ Flask API, model training, and saved model artifacts.
• frontend/ Static single-page UI that calls the API.
• data/ Data conversion and training helper scripts + datasets.
• docs/ Detailed documentation and architecture diagrams.

How It Works

• Training (backend/train_real.py) builds a TF‑IDF + RandomForest model from URL text.
• The Flask API (backend/app_flask.py) loads model.pkl and vectorizer.pkl to score URLs.
• The UI sends a URL to the API and displays the label, score, and reasons.

API

POST /analyze

{ "url": "https://example.com/login" }

Response (example):

{
  "url": "...",
  "label": "SAFE | SUSPICIOUS | MALICIOUS",
  "score": 0-100,
  "reasons": ["..."]
}

Quick Start (Backend)

1. Create and activate a virtual environment (recommended).
2. Install dependencies:
   • pip install -r backend/requirements.txt
3. Run the API:
   • python backend/app_flask.py Note: debug=True is enabled by default for development. Remember to disable it for production.
4. Test:
   • POST http://127.0.0.1:5000/analyze

Train / Refresh the Model

There are two training flows:

• Synthetic data: backend/train.py (generates fake URLs).
• Real data: backend/train_real.py (uses data/openphish_norm.csv and data/benign_norm.csv).

Typical real-data flow:

1. Put raw files in data/:
   • openphish_raw.txt
   • benign_raw.csv (must include a domain column)
2. Run the conversion helper:
   • powershell -File data/prepare_and_train.ps1
   • Add -RunTraining to auto-train after conversion.
3. Training outputs:
   • backend/model.pkl
   • backend/vectorizer.pkl

Frontend

frontend/index.html is a static page that calls a backend URL configured in the script. By default it points to a hosted API (onrender.com). Update the fetch URL if you want to use your local API.

Project Structure

• backend/app_flask.py — Flask API (/analyze)
• backend/train_real.py — TF‑IDF + RandomForest training
• backend/model.pkl / backend/vectorizer.pkl — saved model artifacts
• frontend/index.html — static UI
• data/prepare_and_train.ps1 — data conversion + optional training

Notes / Gotchas

• backend/venv/ is checked in; you can ignore it and create your own environment.