Car Price Prediction (CarCast) - ML Pipeline

A complete machine learning pipeline for predicting car prices using multiple regression models.

Ai2 Project

Project Title: Car Price Prediction Pipeline Objective: Develop a robust regression model to estimate car prices using vehicle features. Methodology: The project implements a full ML pipeline including data cleaning, feature engineering (Target/One-Hot encoding), and a comparative analysis of 5 algorithms (Linear Regression, Logistic Regression, KNN, Decision Tree, K-Means). Outcome: A trained model capable of predicting prices with evaluated accuracy metrics (RMSE, R²).

Overview

This project implements a comprehensive ML workflow that:

• Preprocesses and cleans automotive data
• Engineers features from raw inputs
• Applies multiple encoding strategies for categorical variables
• Trains and compares 5 different machine learning models
• Provides detailed evaluation metrics and visualizations

Models Implemented

Model	Type	Description
Linear Regression	Regression	Baseline linear model for price prediction
Logistic Regression	Classification	Classifies cars into Low/High price categories
K-Nearest Neighbors	Regression	Instance-based learning for price prediction
Decision Tree	Regression	Tree-based model for capturing non-linear patterns
K-Means	Clustering	Unsupervised clustering to find natural groupings

Installation

Prerequisites

• Python 3.8+
• pip package manager

Install Dependencies

pip install pandas numpy scikit-learn matplotlib seaborn xgboost

Dataset

The project requires a CSV file named car_price_prediction.csv with the following columns:

Column	Type	Description
ID	int	Unique identifier
Price	int	Target variable (USD)
Levy	str	Tax levy ('-' or numeric)
Manufacturer	str	Car manufacturer
Model	str	Car model
Prod. year	int	Production year
Category	str	Vehicle category
Leather interior	str	Yes/No
Fuel type	str	Fuel type
Engine volume	str	Engine size (e.g., "2.0 Turbo")
Mileage	str	Mileage with units (e.g., "100000 km")
Cylinders	float	Number of cylinders
Gear box type	str	Transmission type
Drive wheels	str	Drive type (FWD/RWD/AWD)
Doors	str	Number of doors
Wheel	str	Steering wheel position
Color	str	Car color
Airbags	int	Number of airbags

Usage

1. Place your car_price_prediction.csv file in the project directory
2. Open car_price_prediction.ipynb in Jupyter Notebook or VS Code
3. Run all cells sequentially

# Or run via command line
jupyter notebook car_price_prediction.ipynb

Project Structure

car_price_prediction/
├── car_price_prediction.ipynb    # Main notebook
├── car_price_prediction.csv      # Dataset (not included)
└── README.md                     # This file

Pipeline Steps

1. Data Preprocessing

• Clean Levy column (convert '-' to 0)
• Extract numeric values from Engine volume (remove "Turbo")
• Parse Mileage (remove "km" suffix)
• Calculate vehicle Age from production year
• Remove duplicates and outliers (IQR method)

2. Feature Encoding

• One-Hot Encoding: Leather interior, Gear box type, Drive wheels, Wheel
• Target Encoding: Fuel type, Model, Airbags, Cylinders, Manufacturer
• Label Encoding: Category, Color

3. Model Training

• 80/20 train-test split
• StandardScaler for numerical features
• Hyperparameters configurable via Config class

4. Evaluation Metrics

• RMSE (Root Mean Squared Error)
• MAE (Mean Absolute Error)
• R² (Coefficient of Determination)
• MAPE (Mean Absolute Percentage Error)

Configuration

Modify the Config class in the notebook to customize:

class Config:
    TARGET = 'Price'
    TEST_SIZE = 0.2
    RANDOM_STATE = 42
    OUTLIER_REMOVAL = True
    IQR_MULTIPLIER = 1.5

Results

After running the notebook, you'll see:

• Model comparison table sorted by RMSE
• Bar charts comparing RMSE, MAE, and R² across models
• Actual vs. Predicted scatter plot for the best model

License

This project is open source and available for educational purposes.