Neural network training algorithm for carbon dioxide emissions forecast: A performance comparison

Abstract

Artificial neural network with many types of algorithms is known as an efficient tool in forecasting as it is able to handle nonlinearity behaviour of data. This paper investigates the performances of Levenberg-Marquardt and gradient descent algorithms of back propagation neural networks carbon dioxide emissions forecast. The inputs for the model were selected and the ANNs were trained using the Malaysian data of energy use, gross domestic product per capita, population density, combustible renewable and waste and carbon dioxide intensity. The forecasting performances were measured using coefficient of determination, root means square error, mean absolute error, mean absolute percentage error, number of epoch and elapsed time. Comparison between these algorithms show that the Levenberg-Marquardt was outperformed the gradient descent in carbon dioxide emissions forecast.