Product Classification

A machine learning system that classifies products into categories based on their names and brands.

What it does

Takes a product name and brand, predicts which category it belongs to. Built with scikit-learn, PyTorch, and FastAPI.

Train the models:

pip install -r requirements.txt
python run_pipeline.py

Start the API:

python app.py
# Visit http://localhost:8000/docs

Make a prediction:

curl -X POST "http://localhost:8000/classify" \
  -H "Content-Type: application/json" \
  -d '{"product_id": "123", "name": "Chanel No 5 Eau De Parfum", "brand": "Chanel"}'

With Docker:

docker build -t product-classifier .
docker run -p 8000:8000 product-classifier

Project Structure

├── app.py                    # FastAPI application
├── run_pipeline.py           # Train models
├── run_tests.py              # Run tests
├── src/                      # ML code
├── tests/                    # Test suite
├── data/                     # Dataset
├── models/                   # Trained models
├── images/                   # Plots and visualizations
└── docs/                     # Additional documentation

Models

Three models trained on product names and brands:

• Random Forest: 73% test accuracy (best performer)
• PyTorch Neural Network: 71% test accuracy
• Logistic Regression: 49% test accuracy (baseline)

The Random Forest model is used in the API.

API Endpoints

• GET / - API info
• GET /health - Health check
• POST /classify - Classify a single product
• POST /classify/batch - Classify multiple products

Interactive docs at http://localhost:8000/docs

Testing

python run_tests.py

Tech Stack

• scikit-learn, PyTorch
• FastAPI
• pandas, numpy
• Docker

Notes

The dataset is small (~180 training samples, 60 categories), which limits model performance. With more data, accuracy would improve significantly.

See docs/ folder for detailed documentation on the ML pipeline, API usage, and Docker deployment.