Unsupervised Clustering for the Multidimensional Analysis of Rural Energy Poverty in Yucatan, Mexico
摘要
This paper presents the initial phase of a thesis project on energy poverty in rural communities near Mérida, Yucatán. The analysis was based on data from 260 households located in Priority Attention Zones identified in the 2024 Poverty and Social Backwardness Report. These communities are characterized by low-income levels and self-built homes that often lack basic infrastructure and thermal comfort. The survey design combined three established measurement approaches—consensual, expenditure-based, and direct methods—adapted to the rural context of southeastern Mexico. Preliminary findings show that although 96.5% of households are connected to the electrical grid, only 36.7% use LED lighting and 32.8% still rely on firewood for cooking, reflecting persistent and multidimensional energy vulnerability. To analyze these patterns, a data normalization scheme was implemented, followed by PCA and K-Means clustering. This approach identified three typologies of energy poverty: structural exclusion, partial transition, and combined thermal-economic stress. These groupings integrate socio-technical variables such as building materials, appliance ownership, ventilation, energy-use behavior, and perceived cost and comfort. Cluster analysis shows that structurally excluded households (Cluster 1) depend on traditional fuels (0.235); transition households (Cluster 0) use modern cooking (0.776) and show intermediate discomfort (0.628). In contrast, the thermally and economically stressed group (Cluster 2) exhibits higher technological adoption—71% LED lighting and 54% modern cooking—but also displays the greatest thermal discomfort (0.523) and perceived cost burden (0.670). The results provide a multidimensional, data-driven perspective on rural energy deprivation, offering insights for targeted policy and intervention strategies in vulnerable regions.