CAR PRICE PREDICTIon MODEL_

Tolga Unal ___

CAR PRICE PREDICTIon MODEL_

Predicting MSRP using machine learningBuilt with XGBoost & Streamlit Data from carsheet.io

start_

CAR PRICE PREDICTION MODEL

Data Collection

ML Model Building

Project Overview

Key Insights/RW Appl.

Future Work & Concl.

CAR PRICE PREDICTION MODEL

Project Overview

Problem: Car prices vary depending on several factors, including brand, performance, and body type. Buyers and sellers frequently disagree on what a reasonable price is, which causes market mispricing. Objective: Build an accurate model to assist buyers & sellers. Impact:

• Helps dealerships & online marketplaces set fair prices.

• Enables informed decisions for buyers.

• Deployed as a Streamlit web app and Can be integrated into car-selling platforms.

CAR PRICE PREDICTION MODEL

Data Selection and Preparation

Dataset:

• Collected via web scraping from carsheet.io.
• 29,566 rows, 16 columns (from carsheet.io).
• Features: Make, Model, Year, MSRP, Horsepower, Torque, etc.

Data Cleaning:

• Filled missing values (Torque, Horsepower, Cylinders).
• Formatted MSRP (removed $ signs, converted to float).
• Dropped Model, Trim, and Used/New Price

more

CAR PRICE PREDICTION MODEL

Model Building & Evaluation

Models Tested:

• Linear Regression, Random Forest, XGBoost.

Best Model: XGBoost Evaluation Metrics:

• Mean Squared Error (MSE), R-squared (R2), Mean Absolute Error (MAE) and MAPE.

Results:

• Best performance with MAPE = 8.45%.

• Train-test split (80-20) & K-Fold cross-validation.

more

CAR PRICE PREDICTION MODEL

Key Insights

Best Model Performance:XGBoost achieved the lowest error rates:

• MAPE: 8.45%
• MAE: $4,476
• Predictions closely matched the original dataset prices.

Most Important Features: Torque, Horsepower, and Year were the top predictors of car price. Luxury brands (e.g., Rolls Royce, Mclearn) significantly impacted pricing. Data Trends:

• Higher Prices: Large body size, convertible style, twin-turbo engines, automatic transmission, and luxury brands.
• Lower Prices: Compact body size, hatchback body style, and manual transmission.

more

CAR PRICE PREDICTION MODEL

Future Work and Improvements & Conclusion

Enhancements:

• Add price comparison tool.
• Integrate real-time price updates.

Real-World Deployment:

• Partner with dealerships & marketplaces.

Data Expansion:

• Improve accuracy by collecting rare car data.

The machine learning model improves transparency in car pricing, which benefits all market participants.

END

Tolga Unal

THANK YOU_

Predicting MSRP using machine learning Built with XGBoost & Streamlit, Data from carsheet.io

Real-World Application and Impact

Streamlit Web App:

• Developed an interactive web app for price predictions.
• Users input car details & get instant results.

Use Cases:

• Car dealerships: Fair pricing.
• Online marketplaces: Instant estimates.
• Buyers/Sellers: Negotiation insights.

Impact:

• Transparency and accuracy in car pricing
• Make well-informed decisions.

Limitations:

• Historical data, not capture sudden market changes.
• Unique models may not be accurately priced.

GO FOR APP

Feature Engineering & Selection

• One-hot encoding for categorical features.

Feature Selection:

• Pearson correlation heatmap to identify key features.
• Decision Tree model to determine feature importance score.

Key Features: Torque, Horsepower, Year.

Model Optimization

Optimization Process:

• Used Grid Search to find the best learning rate for XGBoost.
• Tested values: [0.01, 0.05, 0.1, 0.3, 0.5, 0.6, 0.8, 1.0].

• Best learning rate: 0.6

Results:

• Before: MAPE = 10.32%, MAE = $5,376 (learning_rate = 0.1).
• After: MAPE = 8.45%, MAE = $4,476 (learning_rate = 0.6).

Impact:

• Improved accuracy and reduced errors.

Real-World Application and Impact

Streamlit Web App:

• Developed an interactive web app for price predictions.
• Users input car details & get instant results.

Use Cases:

• Car dealerships: Fair pricing.
• Online marketplaces: Instant estimates.
• Buyers/Sellers: Negotiation insights.

Impact:

• Transparency and accuracy in car pricing
• Make well-informed decisions.

Limitations:

• Historical data, not capture sudden market changes.
• Unique models may not be accurately priced.

GO FOR APP