A comprehensive data science project analyzing factors that drive used car prices to provide actionable insights for used car dealerships.
This project represents my capstone work for the Professional Certificate in Machine Learning and Artificial Intelligence at UC Berkeley College of Engineering in California. I analyzed 10,000 of the 426,880 used car records to identify key price drivers and provide data-driven recommendations for inventory acquisition and pricing strategies. Using machine learning models, I achieved ~$8,500 RMSE for Ridge Regression. However, after running the full dataset, RMSE for Lasso is only $7, 426.
Since I have studied Linear, Ridge, and Lasso Regression, I decided to focus on these three models to strengthen my skills and gain a deeper intuition about how they perform in practice. To build confidence, I first ran experiments on a 10,000-row sample before scaling up to the full dataset of 426,880 records. One of my biggest challenges was managing the missing values (missingness). This required me to revisit concepts like one-hot encoding and carefully apply them so that categorical variables could be transformed into numerical features the models could use.
After cleaning and preparing the dataset, I compared the three models using cross-validation. The results showed very similar errors, but Lasso Regression slightly outperformed the others on this run:
=== CROSS-VALIDATION COMPARISON (lower RMSE is better) === Model CV_RMSE Best_Params 0 Lasso Regression 7,533.49 {'regressor__alpha': 0.1} 1 Linear Regression 7,533.58 {} 2 Ridge Regression 7,533.70 {'regressor__alpha': 1}
=== BEST MODEL: Lasso Regression ===
- Car age and mileage are the strongest price predictors
- Premium manufacturers command significant price premiums
- Vehicle condition and transmission type significantly impact pricing
- Depreciation rate: ~$2,000-3,000 per year on average
analyzing_car_prices/
├── README.md # Project documentation
├── data/
│ └── vehicles.csv # Original dataset (426K records)
├── Report.md # Detailed analysis report
├── analyzing_car_prices.ipynb # Complete analysis notebook
├── requirements.txt # Python dependencies
├── images/ # Generated charts and visualizations
└── LICENSE # Project license
- Python 3.8+
- pip or conda package manager
-
Clone the repository
git clone <repository-url> cd analyzing_car_prices
-
Install dependencies
pip install -r requirements.txt
-
Launch Jupyter Notebook
jupyter notebook analyzing_car_prices.ipynb
- Best Model: Lasso Regression (slightly better in latest run; Ridge also performed similarly)
- Test RMSE: ~$7,400
- Test MAE: ~$4,900
- Test R²: ~0.73
- Cross-validation: 3-fold CV with hyperparameter tuning
Based on permutation importance analysis:
- Car Age - Most significant factor
- Odometer Reading - Strong negative correlation
- Manufacturer - Premium brands maintain value
- Condition - Excellent condition commands premium
- Transmission Type - Automatic vs manual preference
| Factor | Impact | Recommendation |
|---|---|---|
| Age | -$2,500/year avg | Focus on 3-7 year old vehicles |
| Mileage | Significant drop >100k | Prioritize <60k mile vehicles |
| Condition | 20-40% price variance | Emphasize excellent condition |
| Brand | Premium +30-50% | Target high-value manufacturers |
- Focus on newer vehicles (3-7 years old) for optimal profit margins
- Prioritize low-mileage vehicles (under 60,000 miles) when possible
- Target premium manufacturers that maintain resale value
- Avoid high-mileage vehicles unless priced significantly below market
- Use age as primary pricing factor - expect $2,000-3,000 depreciation per year
- Adjust pricing based on mileage brackets:
- 0-30k miles: Premium pricing
- 30-60k miles: Standard pricing
- 60-100k miles: Moderate discount
- 100k+ miles: Significant discount
- Premium pricing for excellent condition vehicles
- Consider transmission preferences in your local market
- Highlight low mileage as key selling point in marketing
- Emphasize vehicle condition in all marketing materials
- Leverage brand reputation - premium manufacturers justify higher prices
- Be transparent about age impact on pricing
This project follows the CRISP-DM (Cross-Industry Standard Process for Data Mining) framework:
- Business Understanding: Define objectives for used car dealership
- Data Understanding: Explore 426K vehicle records, identify quality issues
- Data Preparation: Clean data, engineer features, handle missing values
- Modeling: Compare multiple ML algorithms, cross-validate, tune hyperparameters
- Evaluation: Assess model performance, extract feature importance
- Deployment: Generate actionable business recommendations
# Data cleaning steps
1. Remove missing prices (target variable)
2. Filter unrealistic prices ($500-$100,000)
3. Clean year data (1990-2023)
4. Clean odometer data (0-500,000 miles)
5. Handle missing categorical data
6. Engineer new features (car_age, manufacturer_grouped)
7. Prepare for modeling (scaling, encoding)- Dataset: Kaggle Used Cars Dataset (subset of 3M records)
- Size: 426,880 records
- Features: 18 columns including price, year, manufacturer, model, condition, etc.
- Target: Price (continuous variable, $500-$100,000 range)
- pandas: Data manipulation and analysis
- numpy: Numerical computing
- scikit-learn: Machine learning algorithms
- matplotlib/seaborn: Data visualization
- jupyter: Interactive development environment
- Modular code structure for reusability
- Comprehensive data validation and cleaning
- Cross-validation with hyperparameter tuning
- Feature importance analysis for interpretability
- Business-focused insights and recommendations
This project was completed as part of my Professional Certificate in Machine Learning and Artificial Intelligence at UC Berkeley College of Engineering in California. It demonstrates practical application of:
- Data Science Methodology (CRISP-DM framework)
- Machine Learning Algorithms (regression, ensemble methods)
- Feature Engineering and data preprocessing
- Model Evaluation and cross-validation techniques
- Business Intelligence and actionable insights generation
This project is licensed under the MIT License - see the LICENSE file for details.
- Fork the repository
- Create a feature branch (
git checkout -b feature/amazing-feature) - Commit your changes (
git commit -m 'Add amazing feature') - Push to the branch (
git push origin feature/amazing-feature) - Open a Pull Request
For questions or collaboration opportunities, please reach out through the repository issues or contact information.
Note: This analysis provides data-driven insights for business decision-making. Always consider market conditions, regional variations, and business constraints when implementing recommendations.