Explaining Regional Energy Poverty in Mediterranean Europe: A Multilevel Regression and Machine Learning Approach

Abstract

Despite the European Union’s extensive efforts to combat energy poverty (EP), especially amid climate change and recent energy crises, several member states continue to face substantial challenges. This paper analyses the determinants of EP in five vulnerable Mediterranean countries (France, Spain, Italy, Greece, and Portugal) using recent regional Eurostat data (NUTS 2, 2024). A multilevel regression framework is complemented by a machine learning analysis, combining fixed-effects logistic models, regularised estimators, and SHAP interpretability, to capture nonlinear interactions and test the robustness of the inferred relationships. The results show that regional economic growth, higher education, better living conditions, and greater labour productivity generally alleviate EP. In addition, inflation, unemployment, and financial instability exacerbated it. Internet access was found to mediate the link between economic growth and EP, but it does not fully explain structural disparities. The interaction between tertiary education and Internet use, although improving connectivity, has been associated with higher utility arrears, reflecting increased energy demand from digital activities. The findings highlight the multidimensional nature of EP and the importance of integrating digital and educational policies into regional development strategies.