Modeling Energy Access Challenges in Europe: A Neural Network Approach to Predicting Household Heating Inadequacy Using Macro-Energy Indicators

Abstract

This study explores the use of machine learning models to predict the percentage of the population unable to keep their houses adequately warm in European countries. The research focuses on applying three machine learning models—ElasticNet, decision trees, and neural networks—using macro-energy indicator data from Eurostat for 27 European countries. Neural networks with Bayesian regularization (BR) achieved the best performance in terms of prediction accuracy, with a regression value of 0.98179, and the lowest root mean squared error (RMSE) of 1.8981. The results demonstrate the superior ability of the BR algorithm to generalize data, outperforming other models like ElasticNet and decision trees, which also provided valuable insights but with lower precision. The findings highlight the potential of machine learning to predict the percentage of the population unable to keep their houses adequately warm, enabling policymakers to allocate resources more efficiently and target vulnerable populations. This research is the result of the application of machine learning models to solve the problem of energy poverty.

Keywords

• energy poverty
• machine learning
• sustainable development
• equitable energy access