Rethinking neighbourhood boundaries for urban planning: a data-driven framework for perception-based delineation

Neighbourhoods are fundamental units for urban analysis, contextualising social and economic life. However, research often relies on administrative geographies as proxies, compromising meaningful delineation for data accessibility. This is a critical problem as administrative boundaries frequently mask significant internal heterogeneity by grouping distinct areas, which can lead to misinterpretations of local conditions and social inequality. This research presents a novel methodology for delineating more homogeneous neighbourhood units based on a large-scale mapping of the urban environment’s visual perception. A deep learning model, trained on street view images and associated perception data, was used to construct a city-wide perception profile of Glasgow, UK. Subsequently, spatial clustering techniques were used to delineate ‘perceptual neighbourhood’ based on shared visual and perceptual characteristics. When compared against standard census geographies using the V-measure, these data-driven neighbourhoods exhibited only moderate spatial association. The results indicate that official census zones, despite their intended socio-economic homogeneity, are often comprised of multiple, distinct perceptual areas. This work provides a scalable, replicable method for generating internally homogeneous, human-centred neighbourhood units. This approach can improve our understanding of the relationship between place and social inequality and inform more targeted, evidence-based urban policy.