Home

📚 GitHub Wiki: Car Dependency Prediction using Urban Form

🧭 Overview

This project explores the spatial prediction of car dependency in England using socio-demographic, socio-economic, and built environment data—including urban form metrics. Machine learning models (Lasso Regression & Random Forest) were applied across 6,791 MSOAs (Middle Super Output Areas) in England.

🏙 Introduction

Urban design influences mobility patterns. This project studies how urban morphology—quantified using geospatial data and momepy—contributes to car dependency in England.

• Understand car dependency using traditional factors.
• Assess urban form as a standalone predictor.
• Analyze combined effects of all factors.

❓ Research Questions

• RQ1: Can socio-demographic, socio-economic, and built environment factors predict car dependency?
• RQ2: Can urban form alone predict car dependency?
• RQ3: When controlling for traditional factors, do urban form metrics significantly improve predictive accuracy?

📊 Data Sources

• Car Ownership: Department for Transport (DFT) – LSOA level – Car license registrations
• Socio-Demographic & Economic: Office for National Statistics – LSOA level – Age, education, income, household composition
• Urban Form: OpenStreetMap + momepy – MSOA level – 68 metrics for buildings, plots, and streets
• Accessibility: National Audit Office (NAO) – LSOA level – Journey times to services via public transport

⚙️ Feature Engineering

• Dependent Variable: Car by population (car_by_pop)
• Independent Variables:
• 27 traditional features (e.g., age, income, education, accessibility)
• 68 urban form metrics (e.g., building elongation, street openness)
• All features standardized and aggregated to MSOA level.

🤖 Modeling

• Techniques: Lasso Regression (interpretable, handles multicollinearity) & Random Forest (non-linear modeling)
• Validation: Spatial K-Fold Cross Validation (K=4) using K-Means clustering
• Evaluation Metric: R² Score

📈 Results

Model	Lasso R²	Random Forest R²
Model 1 (Traditional only)	0.80	0.783
Model 2 (Urban Form only)	0.71	0.76
Model 3 (All Features)	0.81	0.80

• Urban form alone explains ~70% of car dependency variance.
• Urban form adds ~1% predictive power when combined with other features.
• London was consistently the most difficult region to predict accurately.

🏗 Urban Form Insights

Top urban form predictors:

• Cell Alignment (stbCeA): High misalignment = higher car dependency
• Street Openness (sdsSPO): More open streets = lower walkability → higher car usage
• Street Height Profile (sdsSPH): Lower street profile = suburban/rural typology → more car reliance

Urban form reflects architectural styles:

• Modernist/Suburban: Higher car dependency
• Compact/Traditional/Urban: Lower car dependency

✅ Conclusion

• Urban form is a powerful predictor of car dependency.
• It provides additional predictive value beyond traditional socio-economic data.
• Urban planning using morphometrics can inform sustainable transport policy.

🚀 Future Work

• Expand to temporal models (longitudinal prediction).
• Incorporate neural networks or ensemble models while preserving interpretability.
• Extend to other UK regions and integrate vehicle type analysis.
• Study causality in urban form vs. car usage.