A Comparative Analysis Of Linear Econometric And Maschine Learning Approaches To Global Climate-Induced Migration Flows

This study compares linear econometric and machine learning approaches to understanding climate-induced migration from a global perspective, utilizing an expanded dataset covering the period from 1960 to 2020. The analysis contrasts a fixed-effects panel model with a Random Forest machine learning model, each designed to capture the influence of climate factors such as average temperature and precipitation on migration flows using the same underlying data. In the linear model, these climate variables interact with socioeconomic indicators like GDP per capita and agricultural dependency, as well as governance and infrastructure quality, to explain historical migration trends and highlight how institutional and structural resilience can mitigate climate pressures. The Random Forest model, in turn, uncovers non-linear interactions and threshold effects that the linear specification cannot directly assume, yet it confirms the relevance of the chosen variables and interactions by producing a similar level of explanatory power. Together, these approaches show that while climate-related changes significantly shape migration patterns, their impact is highly context-dependent. The findings underscore the advantages of combining traditional econometric modelling with machine learning methods to achieve a more comprehensive understanding of global climate-induced migration flows.