Investigations on machine learning, deep learning, and longitudinal regression methods for global greenhouse gases predictions

Abstract

Combating climate change is one of the key topics and concerns that our community is currently facing these days. Since a few decades ago, greenhouse gases emissions gradually started to increase. Thus, the researchers attempted to find a permanent solution for this challenge. In this paper, different methods of machine learning and deep learning models are applied to evaluate their effectiveness and accuracy in predicting greenhouse gases emissions. To increase the accuracy of the assessment, the data of 101 countries over a period of 31 years (1991–2021) from the official World Bank sources are considered. In this study, therefore, a range of matrices are analyzed including Mean Squared Error, Root Mean Squared Error, Mean Absolute Error, p value, and correlation coefficient for each model. The results demonstrate that machine learning models typically overtake the deep learning models with the support vector regression polynomial model. Besides, the statistical findings of longitudinal regression analysis reveal that by increasing cereal yield, and permanent cropland areas the greenhouse gas emissions are significantly increase (p value = 0.000) and (p value = 0.06) respectively; however, increasing in renewable energy consumption and forest areas will lead to decreasing in greenhouse gas emissions (p value = 0.000) and (p value = 0.07) respectively.