In Environmental science and pollution research international
By considering the existence of two separate analysis families and the usage of different data frequencies, this study aims to examine the effect of method choice, data frequency, and sector-based energy consumption on carbon dioxide (CO2) emissions by performing machine learning (ML) algorithms and time series econometric (TS) models simultaneously. In this situation, the study examines the United States (USA), considers sector-based energy consumption indicators as explanatory variables, uses monthly and yearly data between January 1973 and December 2021, estimates CO2 emissions, and compares the estimation performance of the models. The empirical findings reveal that (i) the ML algorithms outperform the TS models based on R2 and goodness of fit criteria; (ii) the estimation performance of the models increases with the high-frequency (i.e., monthly) data; (iii) the ML algorithms perform much better in case of high-frequency usage; (iv) some thresholds identify the effects of the sector-based energy consumption indicators on the CO2 emissions; (v) electric power and transportation sectors are the most important sectors in the estimation of the CO2 emissions for monthly and yearly data, respectively. Hence, the study provides to help the understanding role of method choice, data frequency, and sector-based energy consumption for the estimation of CO2 emissions. Based on the results, this study proposes that US policymakers should consider the ML algorithms, use higher-frequency data, and include sector-based energy consumption indicators to have a better estimation of CO2 emissions.
Ulussever Talat, Kılıç Depren Serpil, Kartal Mustafa Tevfik, Depren Özer
2023-Feb-25
CO2, Machine learning, Sector-based energy consumption, Time series, USA
