SafePay AI - Fraud Detection System

AI-powered fraud detection system that analyzes SMS messages and transaction patterns to prevent digital payment scams before payment approval.

🎯 Features

• SMS Scam Detection - NLP-based analysis using TF-IDF and Logistic Regression
• Transaction Risk Analysis - ML-based scoring using Random Forest classifier
• Explainable AI - Clear explanations of fraud indicators and risk factors
• Real-time Analysis - Combined risk assessment with actionable recommendations
• Modern UI - Premium dark mode interface with glassmorphism and animations
• Demo Ready - Pre-loaded examples for instant testing

🏗️ Architecture

SafePay AI
├── Backend (Flask + ML)
│   ├── SMS Scam Detector (Logistic Regression + TF-IDF)
│   ├── Transaction Analyzer (Random Forest)
│   └── REST API (/analyze endpoint)
└── Frontend (Vanilla JS)
    ├── Interactive fraud detection form
    ├── Real-time risk visualization
    └── Detailed explanations

🚀 Quick Start

Prerequisites

• Python 3.9 or higher
• pip (Python package manager)

Installation

1. Clone or navigate to the project directory

   cd safepay-ai

2. Create a virtual environment

   # Windows
   python -m venv venv
   venv\Scripts\activate
   
   # Linux/Mac
   python3 -m venv venv
   source venv/bin/activate

3. Install dependencies

   cd backend
   pip install -r requirements.txt

Training the Models

Before running the application, you need to train the ML models:

# From backend directory
cd training

# Train SMS scam detection model
python train_sms_model.py

# Train transaction risk analysis model
python train_transaction_model.py

Expected output:

• Model accuracy metrics
• Feature importance scores
• Saved model files in data/trained_models/

Running the Application

1. Start the Flask backend

   # From backend directory
   python app.py

   Server will start at: http://localhost:5000

2. Open the application

   • Navigate to http://localhost:5000 in your web browser
   • The frontend will automatically connect to the backend API

📊 Demo Scenarios

Click the Quick Examples buttons for instant demos:

1. Scam SMS (High Risk)

• Amount: ₹50,000
• Receiver: New
• Message: Urgent account block warning with suspicious link
• Expected: High risk, BLOCK recommendation

2. Legitimate SMS (Low Risk)

• Amount: ₹599
• Receiver: Known
• Message: OTP verification code
• Expected: Low risk, APPROVE recommendation

3. High Risk Transaction

• Amount: ₹85,000
• Receiver: New
• Message: Prize/lottery scam
• Expected: High risk, BLOCK recommendation

🔍 API Documentation

POST /analyze

Analyze a payment for fraud risk.

Request:

{
  "amount": 50000,
  "is_new_receiver": 1,
  "transactions_today": 1,
  "message": "URGENT: Your account will be blocked..."
}

Response:

{
  "risk_level": "High",
  "risk_score": 85,
  "recommendation": "BLOCK - Manual review required",
  "explanation": "Detailed analysis...",
  "sms_analysis": {
    "is_scam": true,
    "confidence": 92.5,
    "risk_score": 93,
    "indicators": ["Urgency language detected", "Suspicious URL detected"]
  },
  "transaction_analysis": {
    "risk_level": "High Risk",
    "risk_score": 100,
    "confidence": 88.3,
    "risk_factors": ["Very high transaction amount", "New/unknown receiver"]
  }
}

GET /health

Health check endpoint.

Response:

{
  "status": "healthy",
  "message": "SafePay AI is running"
}

🧠 ML Models

SMS Scam Detector

• Algorithm: Logistic Regression
• Features: TF-IDF vectorization (1000 features, bigrams)
• Training Data: 35 labeled SMS samples
• Accuracy: ~85-90%
• Explainability: Keyword detection, URL analysis, urgency patterns

Transaction Analyzer

• Algorithm: Random Forest (100 trees)
• Features: Amount, receiver history, transaction frequency
• Training Data: 50 labeled transactions
• Accuracy: ~85-90%
• Output: 3-class classification (Low/Medium/High risk)

📁 Project Structure

safepay-ai/
├── backend/
│   ├── app.py                      # Flask application
│   ├── config.py                   # Configuration
│   ├── requirements.txt            # Python dependencies
│   ├── models/
│   │   ├── sms_detector.py         # SMS scam detection
│   │   ├── transaction_analyzer.py # Transaction risk analysis
│   │   └── explainer.py            # Model explainability
│   ├── data/
│   │   ├── raw/
│   │   │   ├── sms_dataset.csv     # SMS training data
│   │   │   └── transaction_dataset.csv
│   │   └── trained_models/         # Saved .pkl files
│   └── training/
│       ├── train_sms_model.py      # SMS model trainer
│       └── train_transaction_model.py
└── frontend/
    ├── index.html                  # Main UI
    ├── css/
    │   └── styles.css              # Premium styling
    └── js/
        └── app.js                  # Frontend logic

⚠️ Important Notes

• Demo Only: This is a hackathon project with simulated data
• No Real Integration: No actual UPI/bank APIs are used
• Synthetic Data: All training data is artificially generated
• Not Production Ready: Requires additional security, validation, and testing for real-world use

🎨 Design Features

• Dark mode with vibrant gradients
• Glassmorphism effects
• Smooth animations and transitions
• Responsive design for all devices
• Real-time risk meter visualization
• Interactive demo examples

🛠️ Technologies Used

• Backend: Python, Flask, Flask-CORS
• ML/NLP: scikit-learn, pandas, numpy, nltk
• Frontend: HTML5, CSS3, Vanilla JavaScript
• Design: Custom CSS with modern aesthetics

📝 License

This is a hackathon demo project for educational purposes.

👨‍💻 Development

Built with ❤️ for hackathon demonstration

---

Need Help?

• Ensure Python 3.9+ is installed
• Activate virtual environment before running
• Train models before starting the application
• Check that port 5000 is available