Electricity consumption forecasting in Turkey and Canada: A comparative analysis of machine learning models

Electricity consumption is considered one of the fundamental indicators in energy planning due to its close relationship with economic growth, urbanization, and energy policies. In this study, a comparative analysis is conducted to estimate per capita electricity consumption in Turkey and Canada. Within the scope of the analysis, GDP per capita, industrial value added, the share of renewable energy, electricity transmission and distribution losses, and the urbanization rate are included as independent variables. In addition, a lagged variable is incorporated into the model to capture the temporal dependency of electricity consumption. During the modeling process, Gaussian Process Regression (GPR), Support Vector Regression (SVR), XGBoost, Artificial Neural Networks (MLP), Ensemble Learning methods, and Multiple Linear Regression (MLR) are employed. The findings indicate that machine learning–based methods generally demonstrate high predictive performance. For Turkey, the best results are obtained from the GPR model, whereas for Canada, the Ensemble Learning approach yields superior performance. Furthermore, the MLR model also shows notably strong performance in the Turkey dataset. In addition, the inclusion of the lagged variable is found to improve prediction accuracy. In conclusion, the dynamics of electricity consumption may vary across countries; therefore, a single model cannot be expected to achieve the same level of performance for all cases. The results suggest that machine learning methods provide effective tools for energy planning and policy development. In this context, future studies based on larger datasets and alternative modeling approaches are expected to contribute to the existing literature.