Modeling and predicting provincial per capita electricity consumption in Türkiye: A panel data and machine learning approach

This study provides a province-level assessment of per capita electricity consumption across all 81 provinces of Türkiye for 2015–2023. It employs two-way fixed-effects panel estimation alongside Random Forest modeling to examine regional electricity demand, assess the relative importance of economic, demographic, and climate-related factors, explore potential non-linearities, and evaluate machine-learning evidence in relation to panel-based findings. The results indicate that the industrial share of GDP and cooling degree days (CDD) are positively associated with electricity consumption. By contrast, the enterprise rate is negatively associated with consumption, reflecting differences in provincial economic structure rather than energy-intensive production. Urbanization, renewable energy capacity, and natural gas consumption do not appear to have statistically significant effects within the model. The Random Forest results also support these findings. Prediction error decreases by approximately 22% relative to a baseline model that includes only province and year effects, and variable importance measures highlight the prominent role of the industrial share of GDP. Overall, the findings suggest that provincial electricity demand in Türkiye is more closely related to economic structure and cooling needs than to supply-side factors, with implications for more accurate regional electricity demand forecasting.