🚗 Vehicle Price Prediction using Machine Learning

This project builds an end-to-end machine learning regression system to predict vehicle selling prices based on technical specifications and descriptive attributes.
It demonstrates a complete ML workflow including EDA, preprocessing, feature engineering, model training, evaluation, and prediction.

---

📌 Project Overview

• Problem Type: Regression
• Domain: Automotive / Pricing Analytics
• Goal: Predict vehicle price accurately from structured and unstructured attributes
• Models Used: Random Forest Regressor, LightGBM Regressor
• Best Model: Random Forest Regressor

---

📊 Dataset

• Source: Vehicle listing dataset
• Records: 1,002 vehicles
• Target Variable: price
• Feature Types:
  • Numerical: year, mileage, cylinders, doors
  • Categorical: make, model, fuel, transmission, body, colors
  • Text: description, name

Data Characteristics

• Missing values present in price, mileage, cylinders, colors, description
• Target variable is right-skewed
• High-cardinality categorical features

---

🔍 Exploratory Data Analysis (EDA)

Key insights:

• Newer vehicles generally have higher prices
• Mileage negatively correlates with price
• Vehicles with more cylinders tend to be more expensive
• Fuel type and transmission influence pricing
• Outliers detected in mileage and price

EDA techniques used:

• Histograms
• Scatter plots
• Box plots
• Feature-wise comparisons

---

🔧 Data Preprocessing

• Missing Values

  • Numerical → Median imputation
  • Categorical → Most frequent value
  • Target (price) → Median imputation

• Feature Scaling

  • StandardScaler applied to numerical features

• Encoding

  • OneHotEncoder used for categorical features
  • handle_unknown='ignore' to prevent inference errors

• Pipeline

  • ColumnTransformer + Pipeline used for clean preprocessing

---

🧠 Feature Engineering

New features created:

• Vehicle Age = Current Year − Vehicle Year
• Description Word Count (text complexity signal)
• Make Extraction from vehicle name

Final dataset:

• 2014 features after encoding
• Supports non-linear learning

---

🧪 Model Training

Models Evaluated

• Random Forest Regressor ✅
• LightGBM Regressor

Train-test split:

• 80% Training
• 20% Testing

---

📈 Model Evaluation

🔥 Random Forest Regressor (Best Model)

• MAE: $3,852
• MSE: 44,794,054
• R² Score: 0.82

LightGBM Regressor

• MAE: $5,217
• MSE: 55,458,917
• R² Score: 0.78

➡️ Random Forest outperformed LightGBM across all metrics

---

🔮 Price Prediction

The trained Random Forest model predicts prices for unseen vehicles.

Example Predictions: [31572.38, 40717.47, 40552.83, 54939.23, 47485.55]

---

Predictions represent estimated vehicle selling prices in USD.

---

🛠️ Tech Stack

Tool	Purpose
Python	Programming
Pandas / NumPy	Data processing
Scikit-learn	ML models & pipelines
LightGBM	Gradient boosting
Matplotlib / Seaborn	Visualization
Jupyter Notebook	Experimentation

---

🚀 How to Run

git clone https://github.com/SyedHussain23/Predict_Vehicle_Prices.git
cd Predict_Vehicle_Prices
pip install pandas numpy scikit-learn matplotlib seaborn lightgbm
jupyter notebook Predict_Vehicle_Prices.ipynb

---

🔮 Future Improvements

• Hyperparameter tuning for Random Forest
• Feature importance visualization
• XGBoost / CatBoost comparison
• Outlier handling with robust loss
• Deployment as a price prediction API

---

👨‍💻 Author

Syed Hussain Abdul Hakeem

• LinkedIn: https://www.linkedin.com/in/syed-hussain-abdul-hakeem
• GitHub: https://github.com/SyedHussain23

---

📄 License

This project is open source and available under the MIT License.

---

⭐ Show Your Support

If you found this project useful, consider giving it a ⭐.