17. Detecting and measuring spatial spillover effects and heterogeneity using interpretable tree-based machine learning approaches: an illustration using the Boston housing dataset

This chapter addresses new scientific pathways following the quantitative revolutions in geography and regional science. We observe the rising popularity of artificial intelligence (AI) – specifically machine learning (ML) techniques – in geography and regional science. This chapter sets out to confront the scientific potential and performance of well-known spatial econometric methods with recent achievements in the ML domain. First, it identifies similarities between these two seemingly different methodological paradigms in spatial research. A new spatial ML approach that seeks to integrate the two methods is proposed. As an illustration, tree-based algorithms, such as regression trees, random forest, and extreme gradient boosting in ML are employed, and results are compared to those of a standard spatial Durbin model, with a focus on the detection and measurement of the spatial autoregressive effect. The new methodological approach is demonstrated with some numerical illustrations using the well-known Boston housing dataset.