Exploring built environment correlates of children's mental health and prosocial behaviour using explainable machine learning
File(s)
Accepted version
Author(s)
Zhang, Dongyu
Leng, Hong
Meng, Jiuqi
Liu, Yunzhe
Fecht, Daniela
Type
Journal Article
Abstract
Urban built environments may shape children's mental health and social development, but their combined and potentially nonlinear associations with favourable developmental profiles remain poorly understood. We linked Wave 13 of Understanding Society with multi-source geospatial and street-view data for children aged 8–12 years living in urban areas of England and Wales. The neighbourhood built environment was characterised within 500 m buffers across greenspace, land-use, road-network and service-facility domains. Total Difficulties and Prosocial Behaviour were derived from the Strengths and Difficulties Questionnaire. XGBoost models with SHAP interpretation were used as an exploratory framework to examine built-environment characteristics associated with Normal-category model
outputs, while hierarchical clustering was applied to describe co-occurring environmental configurations. Built-environment characteristics contributed substantially to model interpretation, with greenspace indicators showing prominent contributions and road-network and service-facility characteristics demonstrating outcome-specific nonlinear patterns. Five descriptive neighbourhood typologies were identified, although child-outcome distributions did not differ significantly across clusters. These findings suggest that multidimensional neighbourhood built-environment characteristics may be associated with favourable developmental profiles and highlight the importance of considering environmental configurations and nonlinear relationships in child-friendly urban planning.
outputs, while hierarchical clustering was applied to describe co-occurring environmental configurations. Built-environment characteristics contributed substantially to model interpretation, with greenspace indicators showing prominent contributions and road-network and service-facility characteristics demonstrating outcome-specific nonlinear patterns. Five descriptive neighbourhood typologies were identified, although child-outcome distributions did not differ significantly across clusters. These findings suggest that multidimensional neighbourhood built-environment characteristics may be associated with favourable developmental profiles and highlight the importance of considering environmental configurations and nonlinear relationships in child-friendly urban planning.
Date Acceptance
2026-09-11
Citation
Applied Spatial Analysis and Policy
ISSN
1874-463X
Publisher
Springer
Journal / Book Title
Applied Spatial Analysis and Policy
Copyright Statement
Copyright This paper is embargoed until publication. Once published the Version of Record (VoR) will be available on immediate open access.
License URL
Publication Status
Accepted