ECPR

Install the app

Install this application on your home screen for quick and easy access when you’re on the go.

Just tap Share then “Add to Home Screen”

ECPR

Install the app

Install this application on your home screen for quick and easy access when you’re on the go.

Just tap Share then “Add to Home Screen”

‘Getting Ready’: Using Machine Learning to Forecast Asylum-Related Migration Flows to Greece

European Union
Migration
Public Policy
Quantitative
Big Data
Policy-Making
Refugee
Maria Patouna
University of Oxford
Maria Patouna
University of Oxford

Abstract

Accurately forecasting asylum-related migration trends is crucial for optimising resource allocation and facilitating evidence-based decision-making for border management, reception facilities, integration programs, and humanitarian aid. Nevertheless, its inherent complexity, with multifaceted and non-linear drivers which vary across contexts, poses strains to accurate prediction. This challenge is further compounded by limitations in data availability and quality, including incompleteness, inconsistency, insufficient disaggregation, and inadequate temporal resolution. Consequently, existing forecasting models, while often sophisticated in their statistical design, often lack practical applicability in real-world policy contexts. This paper introduces a novel, pragmatic and policy-oriented approach to forecasting asylum-related migration to Greece, a country significantly impacted by the European migration crisis yet underrepresented in forecasting research. Using machine learning (ML) techniques and a combination of traditional and big data sources, this study develops a forecasting tool to predict first-instance asylum applications from 18 countries-of-citizenship up to three months in advance. An ensemble modelling approach is employed to enhance predictive accuracy. Prioritising real-world applicability, the study exclusively uses data available at the time of prediction or at the required temporal aggregation level. In a departure from currently dominant approaches, the efficacy of demographic forecasts by sex and age group at the sub-national level is explored, challenging the dominance of aggregated country-specific forecasts in the existing literature. The results not only demonstrate that the ML-ensemble significantly outperforms ‘naïve’ models’ predictions, but they also show potential for granular forecasts to rival or even surpass the accuracy of aggregated models. The explained variance of the demographic model is as high as 98% and the out-of-sample performance attains impressive RMSE values as low as 23 to 34 across the three forecasted months. Overall, this study presents a robust AI-powered tool capable of providing timely, demographically detailed forecasts, contributing to the research agendas on predicting international migration flows.