Forecasting CO2 Emissions in Sweden with a Bayesian Neural Network

Abstract: Carbon dioxide (CO2) is the main constituent of greenhouse gases whose increasing concentrations creates a multitude of different environmental problems. Developing an effective predictive modell for forecasting CO2 is therefore of great importance for future policymakers. This thesis aims to develop a Bayesian neural network to forecast CO2 emissions in Sweden from the years 2022-2025. Gross Domestic Product (GDP), renewable energy consumption and oil consumption were used as inputs parameters for the modell. The modell was evaluated with three error metrics namely; the root mean squared error (RMSE), mean squared error (MSE) and mean absolute error (MAE). The results from the evaluation yielded the following values for the error metrics; RMSE = 1.25, MSE = 1.57 and MAE = 0.99 which indicated a low error on the modell’s predictions. The forecasted CO2 emissions steadily decreased from 37 million tonnes in 2022 to approximately 30 million tonnes in 2025. A sensitivity analysis was conducted to examine the impacts of each the input parameter on CO2 emissions, which revealed that oil consumption had the most impact followed by renewable energy consumption and GDP.