Contextualizing the Formation of Individual-Level Migration Intentions across the World Using Machine Learning

Alejandra Rodríguez Sánchez , University of Potsdam
Arne Maaß, University of Potsdam
Jasper Tjaden, Humboldt Universität zu Berlin

Migration intentions are influenced by a wide array of individual-level factors spanning economic, social, demographic, cultural, political, and developmental dimensions. However, the relevance of these drivers is profoundly shaped by the national and regional contexts in which individuals reside. Despite advances in understanding these dynamics, the extent to which the importance of migration predictors varies across countries and regions remains underexplored. Leveraging data from Gallup World Poll (GWP) surveys (2007–2016), we employ supervised and unsupervised machine learning methods, alongside explainable AI techniques, to highlight the role of context in shaping migration intentions. Pooling survey responses across years, we develop an ensemble of ten XGBoost models to predict migration intentions, using most of the core GWP variables as predictors. SHAP values are computed to evaluate feature importance, and dimensionality reduction techniques (PCA and UMAP) are applied to reveal patterns in the salience of predictive factors across countries and regions. Our findings identify a core set of common predictive factors alongside substantial heterogeneity, with regional clustering emerging as a key feature of the formation of migration intentions. These results underscore the importance of country-specific and regional contexts in the predictive strength of migration drivers, suggesting unique geographic dynamics in the formation of migration intentions. By quantifying these variations, our machine learning framework offers new avenues for understanding migration behavior and enhancing the accuracy of migration forecasting models.

See paper

 Presented in Session 48. International Migration and Global Challenges