Investigating Urban Neighborhood Similarity : AI classification of urban neighbourhoods for material-stock recycling — a case study on machine learning vs. vector distance vs. expert knowledge in the city of Vienna

Poster Presentation: Interdisciplinary Workshop – AI in Science and Engineering, Vienna, Austria. https://doi.org/10.34726/12379
Wurzer, Gabriel; Lorenz, Wolfgang; Bindreiter, Stefan
{gabriel.wurzer|wolfgang.lorenz|stefan.bindreiter} (at) tuwien.ac.at
www.dap.tuwien.ac.at/
Vienna; Austria
Keywords: Cityscape Analysis; Artificial Intelligence; Machine Learning; Vector Distances; Similarity Metric; Material Stock; Urban Mining
June, 2026
Abstract.
Vienna holds an enormous, largely unmapped material stock — brick, wood and steel locked inside buildings that will eventually be demolished. To locate where similar recycling potential clusters, we partition the entire city into approximately 16,000 map tiles of roughly 200 m side length and task an artifi cial-intelligence workfl ow with assigning each tile one of fi ve characteristic Viennese building types. Experts (architects and urban planners) manually label only a small set of archetypal tiles; from these few examples, two automatic methods scale the classifi cation to the whole city: a hand-crafted vector-distance similarity and a supervised machine-learning classifi er based on decision trees and random forests. This contribution reports what the AI learned, how the two methods compare in accuracy, where the AI agrees with human experts, and — most revealingly — where and why it does not.





































