Mapping urban change: An explainable machine learning approach to census unit classification and evolution

This is the authors' abstract. We don't add key points for this paper.

Machine learning (ML) and explainable artificial intelligence (XAI) are powerful tools for uncovering complex, multidimensional demographic patterns while ensuring result interpretability, critical for decision-making and policy development. This study introduces a novel ML- and XAI-based methodology for analyzing urban change, adaptable to different locations, datasets, and spatial resolutions. To demonstrate its potential, we applied this methodology to analyze demographic shifts in Madrid and Barcelona (Spain) between the 2011 and 2021 census rounds. By clustering census units using multivariate demographic data, we identified seven clusters with unique distributions and geographic patterns. The analysis reveals significant shifts driven by foreign-born populations, including neighborhood reconfigurations and the expansion of gentrified areas dominated by highly educated young professionals. These trends coincide with the impacts of the 2008–2014 real estate crisis on the decline of family-centric neighborhoods. Additionally, we observe increasing neighborhood homogeneity and citywide segregation. While these shifts challenge social cohesion, they present opportunities for targeted public service provision tailored to demographic needs. This work demonstrates the transformative potential of ML and XAI in urban demographic analysis, paving the way for future studies exploring urban dynamics across different cities and contexts, and enabling more robust, data-driven urban policy decisions.