Prediction of Potential Carbon Dioxide Emissions of Selected Emerging Economies Using Artificial Neural Network

Tackling future global emissions of carbon dioxide is a daunting task. Different black box models have been used to determine the trajectories of CO₂ emissions and other carbon stocks. Trajectories are important because climate modelers use them to project future climate under higher atmospheric CO₂ concentrations. In this paper, fully connected two-layer feed-forward neural network with tangent activation function that comes with hidden neurons as well as linear output neurons was used. The study applied classical nonlinear least squares algorithm such as LM (Levenberg-Marquardt), to predict potential emissions of selected emerging economies. Building the model on the basis of input variables such as crop production, livestock production, trade imports, trade exports, economic growth, renewable and nonrenewable energy consumption. These variables are considered to affect the ecosystems of high rising economic power states. The main idea is to ensure that emerging economies have a clear understanding of expected future emissions so that appropriate measures can be implemented to mitigate its impact. Data for the analysis were obtained from 1971 to 2013 from World Development Indicators and FAOSTAT database. Results indicate an achievement of training performance at epoch 11 when the value of the MSE (Mean Square Error) is 0.0003345 which indicates that the model errors are less than 0.05. Hence, the study concluded that the applied model is capable of predicting potential carbon dioxide emissions in emerging economies with the greatest precision.