Used Car Price Prediction Package

A modular machine learning package for predicting used car prices using XGBoost and Neural Network models.
Supports both training (with GPU) and deployment (CPU, with Streamlit UI and batch prediction).

---

📁 Project Structure

Used_Car_Price/
│
├── src/
│   ├── preprocessing.py         # Data cleaning and preprocessing
│   ├── feature_engineering.py   # Feature engineering for both models
│   ├── xgboost_model.py         # XGBoost model class
│   ├── neural_network.py        # Neural Network model class
│
├── main.py                      # Training pipeline (run on training server)
├── streamlit_app.py             # Streamlit UI for deployment (run on deploy server)
├── generate_fake_cars.py        # Script to generate fake test data
├── requirements.txt             # Python dependencies
├── .gitignore                   # Files and folders to ignore in git
├── readme.md                    # This file
├── models/                      # Saved models (created after training)
└── test data/                   # Test data CSVs for deployment and validation

---

🚦 Server Roles

1. Training Server (with GPU)

• Purpose: Train models, evaluate, and save them for deployment.
• Key Script: main.py
• Outputs:
  • models/xgboost_model.joblib
  • models/neural_network_model.h5

2. Deploy Server (CPU, Local Machine)

• Purpose: Load trained models, predict prices for new data, provide UI and batch prediction.
• Key Script: streamlit_app.py
• Test Data: Test data (CSV) is mandatory and should be placed in the test data/ folder for batch prediction and validation.
• Test Data Generation: Use generate_fake_cars.py if you need synthetic test data.

---

🛠️ Setup & Installation

1. Clone the repository:

   git clone <your_repo_url>
   cd Used_Car_Price

2. Create and activate a virtual environment (recommended):

   python3 -m venv price_analyzer
   source price_analyzer/bin/activate

3. Install dependencies:

   pip install -r requirements.txt

---

🏋️‍♂️ Training Workflow (Training Server)

1. Prepare your training data:
   Place your raw CSV (e.g., vehicles.csv) in the data/ directory.

2. Run the training pipeline:

   python main.py

   • This will preprocess data, engineer features, train both models, evaluate, and save them to models/.

3. Transfer the models/ directory to your deploy server.

---

🚀 Deployment & Prediction (Deploy Server)

1. Ensure the following are present on your deploy server:

   • models/ directory with trained models
   • src/ directory with all modules
   • streamlit_app.py
   • requirements.txt
   • Test data CSVs in the test data/ folder (mandatory for batch prediction)

2. Install dependencies:

   pip install -r requirements.txt

3. Run the Streamlit app:

   streamlit run streamlit_app.py

   • Access the UI in your browser.
   • Enter car details for single prediction or upload a CSV from the test data/ folder for batch prediction (test data required).

---

🧪 Generating Fake Test Data

To create synthetic data for testing or demo:

python generate_fake_cars.py

• This will generate:
  • fake_cars_for_prediction.csv (for batch prediction, no price column)
  • fake_cars_for_validation.csv (with price column for validation)

Move these files into the test data/ folder for use in the deployment UI.

---

📦 Features

• Modular codebase: Clean separation of preprocessing, feature engineering, and modeling.
• Supports both XGBoost and Neural Network models.
• Streamlit UI: For easy local deployment and user-friendly predictions.
• Batch prediction: Upload a CSV from test data/ and get predictions for all entries (test data required).
• Fake data generation: For testing and demonstration.

---

📝 Notes

• Do not retrain models on the deploy server; only use for inference.
• Make sure the input data columns match those expected by the models.
• For best results, use the same preprocessing and feature engineering pipeline for both training and inference.

---

📧 Contact

For questions or contributions, please open an issue or pull request on the repository.