Reconnecting the City; Using Ai To Help Solve Social Isolation In Gothenburg; A Data-Driven, Evidence-Based Approach To Urban Design

Abstract

How can we connect with one another? Social isolation has become a pressing public and urban design issue, especially in rapidly developing and increasingly fragmented cities. In the current era of rapid development in artificial intelligence, the use of interdisciplinary computational methods to support urban design offers a promising research direction. This master’s thesis investigates how AI-supported spatial analysis can inform urban design strategies to address structural social isolation in Gothenburg. The research focuses on Bergsjön, Gothenburg. The methodology combines QGIS-based spatial analysis and Python-based data processing with three statistical modelling techniques: linear regression, logistic regression, and K-means clustering. First, a continuous proxy for structural isolation was constructed based on POI accessibility, combining distance to the nearest facility and the number of facilities within walking distance. Linear regression was then used to examine how angular integration, GFA density relate to isolation patterns. Logistic regression was applied to estimate the probability of each spatial unit belonging to a high isolation-risk condition, producing both a continuous risk map and a categorized design priority map. Finally, K-means clustering was used to classify high-risk areas into different spatial types. The results show that structural isolation in Bergsjön is spatially uneven and strongly related to low integration, weak functional access, and fragmented urban conditions. High-risk areas are often located near green edges, infrastructure boundaries, and spaces with poor spatial continuity. The analysis also shows that different high-risk areas operate through different spatial mechanisms and therefore require different design responses. To translate these findings into site-scale design, City2Graph and Python-based network analysis were used to examine the relationships between paths, buildings, POIs, and network centrality. These analyses helped identify where a community spine should be strengthened and where social nodes should be placed. Based on this evidence, a detailed design proposal is developed for one selected site, focusing on external connectivity, public space as daily infrastructure, and targeted block retrofit. The thesis concludes that AI-supported and graph-based spatial analysis can strengthen urban design by revealing hidden spatial risk patterns, supporting site selection, and linking design decisions more clearly to spatial evidence.