Enhancing CO[formula omitted] emissions prediction in Africa: A novel approach integrating enviroeconomic factors and nature-inspired neural network in the presence of unit root

Abstract

The prediction of CO2 emissions is critical for designing sustainable environmental policies and meeting the Sustainable Development Goals, particularly those related to climate action. Therefore, this paper aims to assess the appropriate predictive model for tracking carbon emissions in four African economies with the highest carbon emissions. The sample countries comprise Algeria, Egypt, Nigeria, and South Africa, spanning the period from 1990 to 2014. Artificial Neural Network coupled with Particle Swarm Optimisation (ANN-PSO) is compared with the Autoregressive model of order 1 (AR), Autoregressive Integrated Moving Average (ARIMA), and Extreme Learning Machine (ELM). Unit root tests are utilised to check the stationarity of the enviroeconomic variables. For the training sample, the ANN-PSO model increased the predictive accuracy of the AR model by 78.50%, 91.18%, 86.4%, and 86.58% for Algeria, Egypt, Nigeria, and South Africa, respectively. For the testing sample, the ANN-PSO model improved the performance of the benchmark model by 95.36%, 83.64%, 97.28%, and 83.03% for Algeria, Egypt, Nigeria, and South Africa, respectively. The evaluation criteria show that ANN-PSO is the most fitting model for predicting carbon emissions in the selected countries. The study concludes that the ANN-PSO model could be valuable for formulating futuristic climate policies to ensure environmental resilience.