Machine learning forecasts suggest a concentration paradox in international student mobility to the United Kingdom

Abstract

Abstract International student mobility is central to higher education finance, soft power, and knowledge production, yet future demand is increasingly uncertain. We analyse Universities and Colleges Admissions Service records on successful international undergraduate applications to the United Kingdom from 86 origin countries between 2010 and 2024, linked to economic and demographic projections, to forecast demand to 2030. A Poisson-based gradient-boosted tree model, a machine-learning method that combines decision trees, is evaluated against autoregressive integrated moving average time-series models and negative binomial gravity models over an out-of-sample period spanning Brexit and the coronavirus pandemic. Forecast errors are lower for the machine-learning model, although gains are modest under structural shocks. Mainland China and Hong Kong, India, and the European Union are projected to account for more than half of international demand by 2030 despite stagnant or declining counts. This concentration paradox increases system vulnerability by intensifying reliance on a narrowing set of origins.