Car Price Prediction – End‑to‑End

This project trains a regression model to predict used car prices using the Kaggle dataset: Vehicle dataset from CarDekho by nehalbirla.

Dataset URL: https://www.kaggle.com/datasets/nehalbirla/vehicle-dataset-from-cardekho

Project Structure

car_price_prediction_app/
├── data/
│   └── car_data.csv                # Put downloaded Kaggle CSV here (rename to car_data.csv)
├── models/
│   └── car_price_pipeline.pkl      # Will be created after training
├── reports/                        # Metrics & plots
├── src/
│   ├── train_regression.py         # Train + evaluate + save pipeline
│   └── utils.py                    # Helpers (plotting, binning)
├── app.py                          # Flask backend + HTML frontend
├── templates/
│   └── index.html                  # Simple web UI for predictions
├── static/
│   └── style.css                   # Minimal CSS
├── streamlit_app.py                # One-file GUI alternative
├── requirements.txt
└── README.md

Quickstart

1. Download the dataset CSV from Kaggle and place it as: ./data/car_data.csv

2. Create a virtual environment and install requirements:

   python -m venv .venv
   .venv\Scripts\activate    # Windows
   source .venv/bin/activate   # macOS/Linux
   pip install -r requirements.txt

3. Train models and generate plots/reports:

   python src/train_regression.py --data ./data/car_data.csv

   Outputs:

   • models/car_price_pipeline.pkl (sklearn Pipeline)
   • reports/metrics.json
   • reports/residuals.png, reports/pred_vs_actual.png, reports/feature_importance.png
   • reports/confusion_matrix_bins.png (classification view on binned prices)

4. Run the Flask app (backend + simple HTML frontend):

   python app.py
   # Open http://127.0.0.1:5000

5. Or use Streamlit GUI:

   streamlit run streamlit_app.py

Notes

• The dataset has columns like: name, year, selling_price (target), km_driven, fuel, seller_type, transmission, owner. The training script infers available columns and handles missing values, encodes categoricals, scales numerics, and performs feature selection (Mutual Information).
• Confusion matrix is not defined for regression; we provide an optional binned-price classification view to visualize mistakes by price range.
• You can toggle models in the script: LinearRegression, RandomForestRegressor, GradientBoostingRegressor, and (optionally) XGBRegressor (requires xgboost).

Repro Tips

• Use --bins 5 (or another integer) to adjust price-range binning for the confusion matrix figure.
• Results will vary depending on preprocessing and random splits.