An Optimal Regularization Neural Network for Historical Mapping of Global CO2 Emissions

Recent studies show that CO2 emission rates will rise dramatically in near future which can ultimately lead to grave climate changes. Artificial neural networks (ANN) are successfully used for forecasting the future trends of various variables. An in-house algorithm is used to train optimal ANN with 5022 training exemplars collected from 162 countries during 1980-2015 and predicted the future emission rates for 2030 and 2050. It is shown that the optimally trained Regularization Network (RN) which has solid roots in multivariate regularization theory, performs more adequately compared to other conventional networks. It is also clearly demonstrated that the optimal stabilization level is essential for filtering the noise and providing faithful generalizations and forecasting. Finally, the most optimal RN is recruited to provide required near and relatively far future forecasts. The predictions indicate that the maximum future emission rates belong to those countries which have both high GDPs and large populations.

Keywords:

Regularization Networks , ANN , CO2 Emissions , Forecasting , Generalization