Phishing Detection & Prevention System

AI/ML-powered URL and email phishing detection

Real-time phishing URL detection · Machine learning classification · Sub-10-second analysis

---

What is this?

A full-stack AI-powered phishing detection system that analyses URLs in real time and classifies them as phishing or legitimate using a trained machine learning model. Built with a FastAPI backend, React frontend, and an ML pipeline trained on a 500,000-URL dataset.

Designed as a practical defensive security tool — the kind a SOC analyst or security engineer would actually use to triage suspicious links.

---

Key Results

Metric	Result
Training dataset size	500,000 URLs
Detection accuracy	85%
End-to-end analysis time	< 10 seconds
False positive rate	Minimised via feature engineering

---

Features

• Real-time URL analysis — paste any URL, get a phishing/legitimate verdict instantly
• ML classification — trained on 500K URLs with feature extraction (URL length, special characters, domain age indicators, subdomain depth, HTTPS presence, suspicious keywords)
• FastAPI backend — RESTful API with clean /predict endpoint
• React frontend — clean UI for URL submission and result display
• Confidence scoring — model outputs probability alongside binary classification

---

Project Structure

Phishing-Detection-Project/
├── backend/                  # FastAPI application
│   ├── main.py               # API endpoints
│   ├── model/                # Trained ML model files
│   └── utils/                # Feature extraction logic
├── ml_model/                 # Model training pipeline
│   ├── train.py              # Training script
│   ├── features.py           # Feature engineering
│   └── evaluate.py           # Model evaluation
├── phishing-detection-frontend/  # React application
│   ├── src/
│   │   ├── App.jsx           # Main component
│   │   └── components/       # UI components
│   └── public/
├── requirements.txt          # Python dependencies
└── .gitignore

---

Tech Stack

Layer	Technology
ML Model	Scikit-learn / Python
Feature Engineering	URL parsing, regex, custom extractors
Backend API	FastAPI (Python)
Frontend	React.js
Dataset	500,000 labelled URLs

---

Installation & Setup

Backend

# Clone the repo
git clone https://github.com/ANIMAALS/Phishing-Detection-Project.git
cd Phishing-Detection-Project

# Install dependencies
pip install -r requirements.txt

# Start FastAPI server
cd backend
uvicorn main:app --reload

API will be running at http://localhost:8000

Frontend

cd phishing-detection-frontend
npm install
npm start

Frontend will be running at http://localhost:3000

---

API Usage

Endpoint: POST /predict

curl -X POST "http://localhost:8000/predict" \
     -H "Content-Type: application/json" \
     -d '{"url": "http://suspicious-login.xyz/paypal/verify"}'

Response:

{
  "url": "http://suspicious-login.xyz/paypal/verify",
  "prediction": "phishing",
  "confidence": 0.94
}

---

How It Works

User submits URL
      ↓
Feature Extraction
  · URL length
  · Special character count (@, -, //)
  · Subdomain depth
  · HTTPS presence
  · Suspicious keyword match
  · Domain structure analysis
      ↓
ML Model Inference
  · Trained on 500K URLs
  · Binary classification
  · Confidence score output
      ↓
Result returned to UI (< 10 seconds)

---

Model Training

The model was trained on a balanced dataset of 500,000 URLs — 250,000 phishing, 250,000 legitimate. Feature engineering extracts 15+ URL-based characteristics without making external DNS or WHOIS calls, keeping inference fast and offline-capable.

# Retrain the model
cd ml_model
python train.py

---

Author

Anirudh N.S. — Cybersecurity Student, Dayananda Sagar University, Bengaluru

Part of a cybersecurity project portfolio alongside WatchDog 2.4 and KeyForge.