Used Car Price Prediction using Ensemble Learning

---

Project Overview

This project predicts the selling price of used cars using Machine Learning regression models.

The workflow covers the complete Machine Learning pipeline, from data preprocessing and feature engineering to training multiple ensemble models and comparing their performance.

The primary objective was to identify the model that provides the best predictive performance while maintaining strong generalization.

---

Dataset

Source: CarDekho Used Cars Dataset

The dataset contains information such as:

• Manufacturing Year
• Fuel Type
• Transmission
• Engine Capacity
• Maximum Power
• Torque
• Seller Type
• Ownership History
• Kilometers Driven

Target Variable

• Selling Price

---

Machine Learning Workflow

• Data Cleaning
• Missing Value Handling
• Feature Engineering
• Exploratory Data Analysis
• Data Preprocessing Pipeline
• One-Hot Encoding
• Train-Test Split
• Model Training
• Cross Validation
• Model Comparison
• Feature Importance Analysis
• Residual Analysis

---

Models Implemented

• Linear Regression
• Decision Tree Regressor
• Random Forest Regressor
• Extra Trees Regressor
• AdaBoost Regressor
• Gradient Boosting Regressor
• Histogram Gradient Boosting Regressor

---

Model Performance

Model	MAE	R² Score	CV Mean
Extra Trees Regressor	₹60,286	0.9775	0.9689
Random Forest Regressor	₹64,710	0.9739	0.9668
HistGradient Boosting	₹73,267	0.9676	0.9613
Gradient Boosting	₹96,606	0.9609	0.9540
Decision Tree	₹77,297	0.9540	0.9568
AdaBoost	₹386,565	0.7221	0.8157
Linear Regression	₹275,656	0.6839	0.6767

---

Best Model

Extra Trees Regressor

Performance

• MAE: ₹60,286
• R² Score: 0.9775
• Cross Validation Mean: 0.9689

The Extra Trees Regressor achieved the highest predictive performance while maintaining excellent generalization across cross-validation folds.

---

Visualizations

Feature Importance

---

Actual vs Predicted Prices

---

Residual Analysis

---

Model Comparison

---

Tech Stack

• Python
• Pandas
• NumPy
• Matplotlib
• Seaborn
• Scikit-learn
• Jupyter Notebook

---

Repository Structure

Used-Car-Price-Prediction/
│
├── notebook/
│   └── used_car_price_prediction.ipynb
│
├── images/
│   ├── feature_importance.png
│   ├── actual_vs_predicted.png
│   ├── residual_plot.png
│   └── model_comparison.png
│
├── requirements.txt
├── README.md
└── LICENSE

---

Key Learnings

• Built reusable preprocessing pipelines.
• Performed feature engineering on real-world automotive data.
• Compared multiple regression algorithms.
• Applied ensemble learning techniques.
• Evaluated models using Cross Validation.
• Interpreted model behavior through feature importance and residual analysis.

---

Future Improvements

• Hyperparameter tuning using RandomizedSearchCV
• SHAP Explainability
• Model deployment using Streamlit
• Prediction API using FastAPI
• Docker containerization

---

If you found this project useful, consider giving it a star!