Forecasting GHG emissions using an optimized artificial neural network model based on correlation and principal component analysis

Abstract

The prediction of GHG emissions is very important due to their negative impacts on climate and global warming. The aim of this study was to develop a model for GHG forecasting emissions at the national level using a new approach based on artificial neural networks (ANN) and broadly available sustainability, economical and industrial indicators acting as inputs. The ANN model architecture and training parameters were optimized, with inputs being selected using correlation analysis and principal component analysis. The developed ANN models were compared with the corresponding multiple linear regression (MLR) model, while an ANN model created using transformed inputs (principal components) was compared with a principal component regression (PCR) model. Since the best results were obtained with the ANN model based on correlation analysis, that particular model was selected for the actual 2011 GHG emissions forecasting. The relative errors of the 2010 GHG emissions predictions were used to adjust the ANN model predictions for 2011, which subsequently resulted in the adjusted 2011 predictions having a MAPE value of only 3.60%. Sensitivity analysis showed that gross inland energy consumption had the highest sensitivity to GHG emissions.