Fake Invoice Detector

AI-powered fake invoice detection system using machine learning to identify fraudulent invoices through text and numerical pattern analysis.

Features

• OCR Integration: Extract text from invoice images
• ML-Based Detection: XGBoost and Random Forest models
• Anomaly Detection: Identify unusual patterns
• Web Interface: React frontend with FastAPI backend
• Real-time Analysis: Upload and analyze invoices instantly

Tech Stack

Backend

• FastAPI: Modern web framework for APIs
• XGBoost: Advanced machine learning algorithm
• scikit-learn: ML utilities and preprocessing
• TextBlob: Natural language processing
• pandas: Data manipulation
• Tesseract OCR: Text extraction from images

Frontend

• React: User interface
• Material-UI: Modern UI components
• Axios: API communication

Installation & Setup

Quick Setup

#Clone the repository
git clone https://github.com/shreeyashree-65/fake-invoice-detector.git
cd fake-invoice-detector

#Run automatic setup
python setup.py setup

#Start development servers
python setup.py dev

Manual Setup

Backend Setup

cd backend
pip install -r requirements.txt
python src/data_generator.py  # Generate training data
python src/model_trainer.py   # Train ML models
python app.py                 # Start FastAPI server

Frontend Setup

cd frontend
npm install
npm start

Docker Setup

#Build and run with Docker
docker-compose up --build

#Or use the setup script
python setup.py docker-build
python setup.py docker-start

OCR Setup

For image processing capabilities:

1. Download Tesseract from: https://github.com/UB-Mannheim/tesseract/wiki
2. Install and add to PATH
3. Update pytesseract.pytesseract.tesseract_cmd in ocr_processor.py

Machine Learning Approach

Feature Engineering

• Text Features: Vendor name patterns, description analysis
• Numerical Features: Amount patterns, tax calculations
• Anomaly Detection: Isolation Forest for unusual patterns

Models Used

• XGBoost: Primary classifier for fraud detection
• Random Forest: Ensemble method for robustness
• Isolation Forest: Anomaly detection for unknown patterns

Training Data

• Synthetic invoice data with realistic patterns
• 50% genuine invoices with proper formatting
• 50% fake invoices with suspicious indicators

Usage

1. Upload Invoice: Drag and drop invoice image
2. OCR Processing: Text extraction from image
3. Feature Extraction: Analyze text and numerical patterns
4. ML Prediction: Classify as genuine or fake
5. Results: View prediction with confidence score

License

This project is licensed under the MIT License - see the LICENSE file for details.

Author

SHREEYA P S - shreeyashree-65