Abstract
This study develops a robust methodological framework for measuring and analyzing multidimensional energy poverty using unit-level household survey data. The approach integrates logistic principal component analysis (Logistic PCA) to construct a composite index that assigns unequal weights to diverse energy deprivation indicators, thereby capturing the heterogeneity and complexity of energy poverty more accurately than equal-weight methods. The index is further disaggregated into moderate and severe categories, enabling a nuanced assessment of deprivation intensity. To complement the measurement stage, artificial intelligence techniques specifically multilayer perceptron (MLP) and artificial neural networks (ANN) are employed to model the socio-demographic and economic determinants of energy poverty. This dual-stage design allows for both explanatory and predictive insights: the statistical modeling validates the significance of key predictors such as household wealth, family size, and access to basic amenities, while the AI models enhance predictive accuracy for identifying high-risk households and regions. By combining unequal-weight composite measurement with AI-driven predictive modeling, the framework offers a scalable and transferable tool for researchers and policymakers. It facilitates targeted, data-driven interventions aimed at reducing energy poverty and promoting equitable energy access. The methodological innovations presented here are adaptable to diverse contexts, making them valuable for comparative studies and policy applications beyond the specific dataset used.