🚗 Pre-Owned Vehicle Price Estimation 🚗

With Random Forest Regression

---

📌 Project Overview

With thousands of used cars listed on platforms like CarDekho, it's difficult for sellers to know the right asking price. Pricing too high means no buyers; pricing too low means losing money.

This project builds a Random Forest Regression model that predicts the resale price of a pre-owned vehicle based on key features like engine size, max power, mileage, fuel type, and more — giving sellers a data-driven price suggestion based on real market conditions.

---

📁 Repository Structure

Resale-Car-Price-Prediction/
├── images/
│   ├── selling_price_distribution.png
│   ├── correlation_heatmap.png
│   └── model_comparison.png
│
├── Resale_Car_Prediction.ipynb
├── cardekho_imputated.csv
├── car_price_predictor.pkl
├── preprocessor.pkl
├── .gitignore
└── README.md

---

🛠️ Libraries Used

• scikit-learn
• pandas
• numpy
• matplotlib
• seaborn
• joblib

---

📊 Dataset

The dataset is scraped from cardekho.com

• Samples: 15,411
• Features: 13
• Target Variable: selling_price — resale price in INR

Feature	Description
model	Car model name
vehicle_age	Age of the vehicle (in years)
km_driven	Total kilometers driven
seller_type	Individual / Dealer / Trustmark Dealer
fuel_type	Petrol / Diesel / CNG / LPG
transmission_type	Manual / Automatic
mileage	Fuel efficiency (km/l)
engine	Engine displacement (cc)
max_power	Maximum power (bhp)
seats	Number of seats

---

⚙️ Workflow

1. Load the dataset from cardekho_imputated.csv
2. Check for missing values — dataset is pre-imputed, no nulls found
3. Drop irrelevant features — car_name and brand (redundant with model)
4. Identify feature types — 7 numerical, 4 categorical
5. Exploratory Data Analysis — price distribution and correlation heatmap
6. Apply Label Encoding on model column (120 unique values)
7. Train-Test Split — 80% train / 20% test (random_state=42)
8. Preprocessing — OneHotEncoder for categorical, StandardScaler for numerical via ColumnTransformer
9. Train and compare 6 models — Linear Regression, Lasso, Ridge, KNN, Decision Tree, Random Forest
10. Hyperparameter tuning using RandomizedSearchCV with 100 iterations and 3-fold CV
11. Evaluate using RMSE, MAE, and R² Score
12. Save model and preprocessor using joblib

---

📈 Exploratory Data Analysis

 

Key Insights:

• Most cars are priced between ₹2–15 lakhs — heavily right-skewed distribution
• max_power has the strongest positive correlation with selling price (0.75)
• engine is the second strongest predictor (0.59)
• vehicle_age and km_driven show negative correlation with price as expected
• mileage negatively correlates with engine size — bigger engines are less fuel efficient

---

🤖 Model Comparison

Model	Train R²	Test R²
Linear Regression	0.6218	0.6645
Lasso	0.6218	0.6645
Ridge	0.6218	0.6645
K-Neighbors Regressor	0.8691	0.9150
Decision Tree	0.9995	0.8823
Random Forest	0.9791	0.9303

Random Forest was selected as the final model — best test R² with no overfitting.

---

🔧 Hyperparameter Tuning

rf_params = {
    "max_depth"        : [5, 8, 15, None, 10],
    "max_features"     : [5, 7, "sqrt", 8],
    "min_samples_split": [2, 8, 15, 20],
    "n_estimators"     : [100, 200, 500, 1000]
}

Tuning method: RandomizedSearchCV — n_iter=100, cv=3, n_jobs=-1

Best Parameters:

RandomForestRegressor(
    n_estimators=1000,
    min_samples_split=2,
    max_features=7,
    max_depth=None
)

---

📉 Final Model Results

Metric	Train	Test
R² Score	0.9804	0.9403
RMSE	₹1,26,209	₹2,12,015
MAE	₹38,890	₹98,050

---

🧪 Sample Prediction

import joblib
import pandas as pd

rf_model = joblib.load('car_price_predictor.pkl')
preprocessor = joblib.load('preprocessor.pkl')

# Sample car — Ford Ecosport
sample = pd.DataFrame([{
    'model'            : 38,        # label encoded value for Ecosport
    'vehicle_age'      : 6,
    'km_driven'        : 30000,
    'seller_type'      : 'Dealer',
    'fuel_type'        : 'Diesel',
    'transmission_type': 'Manual',
    'mileage'          : 22.77,
    'engine'           : 1498,
    'max_power'        : 98.59,
    'seats'            : 5
}])

sample_transformed = preprocessor.transform(sample)
predicted_price = rf_model.predict(sample_transformed)
print(f"Predicted Selling Price: ₹ {predicted_price[0]:,.0f}")
# Output: Predicted Selling Price: ₹ 6,08,500

---

🚀 How to Run

1. Clone the repo

git clone https://github.com/AnmolPatel20/Resale-Car-Price-Prediction.git
cd Resale-Car-Price-Prediction

2. Install dependencies

pip install pandas numpy matplotlib seaborn scikit-learn joblib

3. Run the notebook

jupyter notebook Resale_Car_Prediction.ipynb

---

📌 Notes

• Both car_price_predictor.pkl and preprocessor.pkl must be loaded together for prediction
• The preprocessor handles all encoding and scaling — never pass raw data directly to the model
• The model column requires the label encoded integer value, not the string name

---

🙋 About

I'm on my machine learning journey — building, experimenting and documenting as I go. Every notebook here represents something I've genuinely tried to understand, not just run. 🚀

• GitHub: @AnmolPatel20
• Portfolio: anmolpatel20.github.io/Anmol_Portfolio

🙏 Acknowledgements

Thanks to Krish Naik Sir whose Udemy course has been a great resource throughout this learning journey.

"Not all those who wander are lost." — J.R.R. Tolkien

---

⭐ Star this repo if you found it helpful!