Forecasting carbon emissions in Turkey’s energy production using a regression-based hybrid LSTM-GRU fusion model

Carbon emissions are a major driver of global warming and climate change, making their prediction essential for sustainable energy planning. Forecasting future carbon emission trends supports both environmental impact reduction and long-term policy development. Since fossil fuel–based electricity generation is one of the largest contributors to global emissions, accurately modeling these trends is crucial for guiding renewable energy investments and emission-reduction strategies. This study conducts a time-series analysis of fossil fuel–based electricity production to forecast carbon emissions. Deep learning forecasting methods, including Long Short-Term Memory (LSTM), the Gated Recurrent Unit (GRU), and a hybrid LSTM–GRU architecture, were employed to capture temporal patterns in the data. Feature engineering techniques such as lagged variables, moving averages, and seasonality indicators were incorporated to enhance predictive performance. Additionally, regression analysis was used to represent the linear relationship between electricity production and carbon emissions. Model accuracy was evaluated using standard error metrics, including the Mean Absolute Error (MAE), Mean Square Error (MSE), Root Mean Square Error (RMSE), and the coefficient of determination (R²). The findings reveal that the hybrid model achieved superior predictive accuracy across all metrics and demonstrated strong correlation with fossil-based electricity trends. These results provide valuable insights for policymakers seeking to develop effective low-carbon energy strategies and climate mitigation plans.