Graph-based machine learning in urban form characterization: a critical methodological review with structured taxonomy and methodological framework
Year: 2026
Type: article
Study design: Systematic Review
Abstract: Graph-based machine learning offers a powerful paradigm for encoding the relational structure of urban form, yet its methodological foundations remain fragmented across disciplinary boundaries. This study presents the first scoping review dedicated to graph-based machine learning for intrinsic urban form characterization, focusing on methods that operate directly on urban topology rather than on functional prediction tasks. Following the PRISMA framework, the review synthesizes 24 and develops a four-axis taxonomy covering urban form elements, graph construction strategies, learning paradigms, and morphological applications, alongside a reproducible seven-stage methodological framework. The central finding is that analytical performance depends less on model architecture than on representational choices made upstream: the abstraction of cities into nodes and edges functions as an epistemic filter that determines which morphological patterns become computationally learnable. While current approaches effectively discover latent typologies within spatially continuous fabrics, they systematically overlook conditions of urban fragmentation, leaving cities shaped by armed conflict, militarized partitioning, and sustained destruction analytically invisible. This study argues for extending graph representations to treat spatial rupture not as noise but as a constitutive feature of urban form, repositioning representation itself as a central theoretical and methodological concern in graph-based urban morphology.
Source: Journal of Geographical Systems
Language: English
Subfield: Building and Construction
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,490
OpenAlex ID: W7214454744