Exploring the complex origins of energy poverty in The Netherlands with machine learning

Abstract

Energy poverty is receiving increased attention in developed countries like the Netherlands. Although it only affects a relatively small share of the population, it constitutes a stern challenge that is hard to quantify and monitor, hence difficult to effectively tackle through adequate policy measures. In this paper we introduce a framework to categorize energy poverty risk based on income and energy expenditure. We propose the use of a machine learning classifier to predict energy poverty risk from a broad set of socio-economic parameters: house value, ownership and age, household size, and average population density. While income remains the single most important predictor, we find that the inclusion of these additional socio-economic features is indispensable in order to achieve high prediction reliability. This result forms an indication of the complex nature of the mechanisms underlying energy poverty. Our findings are valid at different geographical scales, i.e. both for single households and for entire neighborhoods. Extensive sensitivity analysis shows that our results are independent of the precise position of risk category boundaries. The outcomes of our study indicate that machine learning could be used as an effective means to monitor energy poverty, and assist the design and implementation of appropriate policy measures.