Using OpenStreetMap, Census, and Survey Data to Predict Interethnic Group Relations in Belgium: A Machine Learning Approach.

The article investigates how local neighborhood "spaces of encounter"—defined as public and semi-public places where residents of different ethnic backgrounds may interact—predict intergroup relations, specifically perceived ethnic threat and intergroup friendship, in Belgium. Using georeferenced survey data from the 2020 Belgian National Election Study linked with spatial features from OpenStreetMap (OSM) and census-based neighborhood indicators, the study employs supervised machine learning to assess the predictive importance of various types of spaces alongside traditional neighborhood ethnic diversity, socioeconomic status, and individual characteristics. Results indicate that educational, functional, public open, and user-selecting spaces are important predictors of intergroup friendship, while functional, third, retail, and natural spaces relate more to perceived ethnic threat; however, individual sociodemographic factors and neighborhood ethnic diversity and disadvantage have stronger predictive power overall. The study introduces a novel typology of spaces of encounter operationalized through OSM data and highlights the potential and limitations of using such geospatial data to understand neighborhood-level interethnic dynamics, suggesting further research incorporating actual usage and perceptions of these spaces.