Characterizing the global greenhouse gases emissions using machine learning

Considering the current context, in which sustainable economic growth is sought with an emphasis on policies and incentives associated with environmental issues, this study investigated the relative importance of socioeconomic determinants in understanding the profile of greenhouse gas emissions using a "machine learning" approach. A "random forest" model was estimated using data on economic productive capacity and the amount of greenhouse gas emissions between 1990 and 2018. The sample studied consisted of countries representing the largest and smallest global economies, selected based on their level of economic activity during the period. Initially, the most relevant variables were identified using the recursive variable elimination technique; then, the model was trained using the "cross-validation" technique; and finally, it was validated with the data selected for testing. The performance metrics did not indicate overfitting problems, and the residuals of the estimates behaved according to the normal distribution.Based on the model estimated in this work, it was observed that the profile of greenhouse gas emissions was influenced differently depending on the country analyzed, such that the more or less relevant factors appeared to be associated with the level of economic activity. Thus, the discussions and modeling presented in this work aimed to encourage incentive policies and control measures directed at the most relevant sectors, which could contribute to sustainable economic growth.