Identifying and Describing Energy-Poor Household Groups. A Comparison Between Two Different Methods: Conventional Statistical Characterisation and Artificial Intelligence-Driven Clusterisation

In recent years, numerous indicators for measuring energy poverty have been identified. Some of them are used officially or have a certain standardisation vocation, being relatively common among energy poverty studies for the identification of the phenomenon. This identification is an essential element for a first approach to the problem, but it is necessary to take a further step in characterising the phenomenon. This step necessarily involves characterising the households that suffer energy poverty. Identifying what these households are like, what are the characteristics that define them and the features that make a household more at risk of suffering energy poverty, and whether there are different types of households and whether they can be grouped together, are some of the questions that continue to be relevant in the study of energy poverty. This research compares two different methods of approaching the above-mentioned questions, starting from an analysis based on the income and expenditure approach. In the first, more traditional method, the characterisation of these households is carried out using conventional statistical tools, allowing a general identification of the most prevalent characteristics of households in energy poverty belonging to different vulnerability groups. In the second, artificial intelligence techniques are used to go a step further, not only characterising households but also subdividing the vulnerability groups to which they belong, identifying common characteristics that go beyond those defining the energy poverty phenomenon. The use of artificial intelligence in the study of energy poverty, by unravelling the specific characteristics and needs of the different subgroups affected by the phenomenon, can enable the personalisation of the construction solutions applied to the housing stock in which these households live. This favours a better coverage of their needs, with greater cost efficiency and results that are better adjusted to the initial conditions, both sociodemographic and constructive.