Understanding and alleviating energy poverty is critical for sustainable development. This study harnesses a suite of Machine Learning (ML) algorithms to predict Multidimensional Energy Poverty Index (MEPI) and to highlight the spatial distribution of energy poverty. We assess the predictive accuracy of Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), Multiple Linear Regression (MLR), and XGBoost models. The RF model outperforms others, achieving an R2 value of 0.92 and a Pearson Correlation Coefficient (PCC) of 0.97 on the testing dataset, indicative of a highly accurate prediction capability. XGBoost also demonstrates strong predictive power with corresponding values of 0.88 and 0.94, respectively. Our spatial analysis, revealing significant clustering of energy poverty with a Global Moran’s I value of 150.39, indicates that energy poverty is not only geographically concentrated but also intricately linked to socio-economic factors such as income levels, access to education, and nutritional status. These insights underscore the necessity of region-specific and socio-economically informed policy interventions. The results inform targeted interventions, particularly highlighting the critical roles of education and nutrition in mitigating energy poverty. The RF model’s accuracy rate of 92% on the testing set suggests that improvements in these sectors could significantly influence MEPI scores. The integration of ML and spatial analysis offers a nuanced and actionable understanding of energy poverty, paving the way for targeted, evidence-based policy formulation aimed at achieving SDG7: ensuring access to affordable, reliable, sustainable, and modern energy for all.