Projects

Urban Mobility Index

PythonTensorFlowGISJavaScript
Spring 2022
Exploring Urban Data with Machine Learning
Columbia GSAPP

An ML model trained on 600+ neighborhoods predicts walkability from street connectivity and transit density — giving planners a score for any location, even without pre-calculated metrics.

"The model reaches 84% accuracy — street connectivity alone explains more walkability variance than transit proximity."

01

Approach

Method
Machine Learning
Walkscore API
Street network analysis

The project addresses urban dependency on vehicular transportation by developing a tool to assess city walkability. Using Walkscore.com datasets, we built a machine learning model that predicts urban mobility efficiency based on street connectivity and transit density.

The interactive platform enables urban planners — especially those lacking pre-calculated metrics or data processing capacity — to evaluate neighborhood walkability scores for any location.

02

Results

  • Street connectivity is the strongest feature. Intersection density and block size explain more walkability variance than transit stop count or distance to rail.
  • The model generalizes across city types. Trained on mixed dense and suburban neighborhoods, the model performs consistently across Northeast, Midwest, and West Coast metros.
  • Planners can score any address. The interactive tool lets users input coordinates and get a predicted walkability score with feature importance breakdown — useful for sites without existing Walkscore coverage.

Team: Kirthi Balakrishnan, Kit Nga Chou, Lizzie Lee, Michelle Chen

Course by Professor Boyeong Hong