Forecasting the annual carbon dioxide emissions of Malaysia using Lasso-GMDH neural network-based

Abstract

In this study, it was intended to develop an accurate forecasting model for the annually CO2 emission of Malaysia in the short-term. For this purpose, the Group Method of Data Handling (GMDH) model as one of the Neural Networks (NNs) was utilized to structure a nonlinear time-series based forecasting model. In order to improve GMDH prediction accuracy, this paper highlights the drawbacks of using the least square method to solve model parameters and attempts to use the Lasso method (Lasso-GMDH). A case study with the proposed model was carried out for one-year-ahead forecasting of CO2 emissions data during the years 2000-2016. Three different models: grey model GM(1,N), artificial neural network (ANN) and GMDH models were investigated to model the Co2 emission forecast. The comparison revealed that Lasso-GMDH model has the highest general performance for forecasting the annually CO2 emission of Malaysia.