A data-driven regression model predicting the resale price of used Ford vehicles using Linear Regression. Achieves R² of 0.840, explaining 84% of price variance with interpretable feature importance.
Real-world machine learning solving the used car pricing challenge
End-to-end automated workflow from raw data to predictions
┌─────────────────────────────────┐ │ Raw Data (ford.csv) │ │ 17,966 rows × 9 columns │ └─────────┬───────────────────────┘ │ ▼ ┌─────────────────────────────────┐ │ Data Inspection & Quality │ │ • No missing values │ │ • 154 duplicates identified │ └─────────┬───────────────────────┘ │ ▼ ┌─────────────────────────────────┐ │ Exploratory Data Analysis │ │ • 22 visualizations │ │ • Correlation analysis │ └─────────┬───────────────────────┘ │ ▼ ┌─────────────────────────────────┐ │ Feature Engineering │ │ • One-Hot Encoding (37 cols) │ │ • StandardScaler (5 features) │ └─────────┬───────────────────────┘ │ ▼ ┌─────────────────────────────────┐ │ Train / Test Split │ │ 67% Train | 33% Test │ └─────────┬───────────────────────┘ │ ▼ ┌─────────────────────────────────┐ │ Linear Regression Training │ │ Training time: < 100ms │ └─────────┬───────────────────────┘ │ ▼ ┌─────────────────────────────────┐ │ Model Evaluation │ │ • R² = 0.840 │ │ • RMSE = £1,900 │ │ • 5-Fold CV (stable) │ └─────────┬───────────────────────┘ │ ▼ ┌─────────────────────────────────┐ │ Model Saved (joblib) │ │ ford_price_model.pkl │ └─────────┬───────────────────────┘ │ ▼ ┌─────────────────────────────────┐ │ Prediction / Inference │ │ New car → Price (GBP) │ └─────────────────────────────────┘
22 professional visualizations covering distributions, correlations, outliers, and feature relationships. Every insight is data-driven.
R² of 0.840 on test set with RMSE of £1,900. Cross-validation confirms the model is stable and generalises well.
Feature importance analysis reveals that year and mileage are the top price drivers, with clear business insights.
5-Fold cross-validation with standard deviation < 0.01 proves the model doesn't overfit to any particular data split.
Complete pipeline with saved model (joblib), reproducible code, and clear documentation. Ready for deployment.
Training completes in under 100ms. Inference takes less than 1ms per prediction. Scalable for real-time applications.
Faster, more consistent trade-in appraisals. Remove human guesswork and provide objective pricing based on 17,966 real market transactions.
Confidence that your listing price is fair and competitive. Understand which features actually drive value in the used Ford market.
Dive into the data, explore 22 visualizations, examine model performance, or try making your own price predictions.