Graph-based machine learning in urban form characterization: a critical methodological review with structured taxonomy and methodological framework
Open Access
- 26 September 2026
- journal article
- research article
- Published by Springer Nature in Journal of Geographical Systems
- p. 1-40
Abstract
Graph-based machine learning (graph-based ML) offers a structurally well-suited paradigm for encoding the relational properties of urban form, yet its application to urban morphological characterization remains fragmented across disciplinary boundaries. This study presents a critical methodological review of graph-based ML for intrinsic urban form characterization, synthesizing 24 studies through a transparent, reproducible search, screening, and extraction process. The review develops a four-axis taxonomy covering urban form elements, graph construction strategies, learning paradigms, and morphological applications, alongside a reproducible seven-stage methodological framework. Grounded in established principles of representation learning, the review advances the argument that upstream representational choices are a consequential and insufficiently examined source of variation in graph-based urban morphology, with the abstraction of cities into nodes and edges shaping which morphological patterns become computationally accessible. Across the 24 reviewed studies, within-study comparisons show that graph construction strategy and morphometric attribute selection can substantially alter analytical outcomes, a pattern recurring across unsupervised, supervised, and semi-supervised paradigms; the corpus does not, however, contain studies that vary representation and architecture systematically enough to establish which factor contributes more to outcome variation overall. The review further identifies graph construction strategy selection as following from three jointly operating determinants (analytical scale, theoretical conception of meaningful adjacency, and task type), a mechanism explaining representational diversity without reducing it to arbitrary methodological preference. A critical gap is identified: current construction strategies provide limited explicit encoding of spatial discontinuity — its causes, permeability, temporal status, and morphological meaning — constraining analysis of fragmented urban conditions. Three tractable future directions are proposed to address this limitation; these extend beyond the reviewed evidence rather than emerging from the corpus as findings of equivalent standing.Keywords
All Related Versions
Funding Information
- Queensland University of Technology
This publication has 52 references indexed in Scilit:
- From buildings to blocks: Graph contrastive learning for integrated urban architectural morphology classificationFrontiers of Architectural Research, 2026
- Detection and clustering of urban form types with machine learning: insights into Thessaloniki's urban planning and evolutionComputational Urban Science, 2025
- Modelling Road Networks with Node Structural Features and Graph Convolutional Networks2025
- From grids to dendrites: Quantifying spatial heterogeneity in urban road networksComputers, Environment and Urban Systems, 2025
- A New Approach for Road Type Classification Using Multi-stage Graph Embedding Method2023
- Data analytics of urban fabric metrics for smart citiesFuture Generation Computer Systems, 2020
- The Handbook Of Urban Morphology2017
- A machine learning-based method for the large-scale evaluation of the qualities of the urban environmentComputers, Environment and Urban Systems, 2017
- OSMnx: New methods for acquiring, constructing, analyzing, and visualizing complex street networksComputers, Environment and Urban Systems, 2017
- Urban Morphology2016