Land use is recorded, but slowly and unevenly. This project asks whether built-form morphology, environmental data and proximity to city features carry enough signal to predict it, and how much data a model needs before it starts being right.
Context
IAAC · Barcelona
Year
2026
Stack
Machine learning · Python · GIS · Urban data
Brief
NYC – Urban Land Use investigates whether machine learning can classify urban areas as commercial or residential using built-form morphology, environmental data, and proximity to surrounding city features. The project reframes land-use prediction as a spatial grid-classification problem, combining expanded urban datasets to identify the conditions required for accurate prediction at the city scale.
“Can we predict the land use of a space based on existing environmental information from official and unofficial sources?”
What’s the sweet spot for a Machine Learning model to start predicting accurately at an urban level?
01 · Question
The question is whether official and unofficial environmental information is enough to say what a space is used for. Built-form morphology, environmental readings and proximity to surrounding city features are the only inputs; the label is what has to be inferred.
02 · Method
Reframing the city as a spatial grid turns an ambiguous urban question into a classification task with a clear unit of analysis. Expanded urban datasets are joined onto that grid, so every cell carries the same feature vocabulary.
03 · Findings
The useful result is not a single accuracy figure but a threshold: how much data, at what resolution, before prediction becomes trustworthy at city scale. Below it the model is confidently wrong, which is the failure mode that matters.
