Estimating Functional Urban Areas without Commuting Data

The number of people living in the commuting zones of large cities currently accounts for almost a fifth of the global population. However, these populations remain poorly represented in global urban datasets due to the widespread scarcity of commuting data. In this project, I replicated and adapted the machine-learning approach of Moreno-Monroy et al. (2021) to estimate Functional Urban Areas (FUAs) worldwide using only globally available geospatial indicators.

I trained a logistic regression model on EU–OECD FUAs using distance to the city center, accessibility, Urban Center population, population density, and sub-national Human Development Index (sHDI). The final model achieved an accuracy of 0.71 and a sensitivity of 0.69 on held-out FUAs, producing plausible commuting zones across a wide range of cities.