Sara Ghalehnovi - M.Sc. StudentDepartment of Civil Engineering, Ferdowsi University ofMashhad, Mashhad, Iran
Afsaneh Ghalehnovi - Assistant ProfessorDepartment of Architecture, Khorasan Institute of HigherEducation, Mashhad, Iran

China, India, and the United States are among thecountries with the highest energy consumption and greenhouse gasemissions, including CO۲, on a global scale. Iran also produces alarge amount of carbon dioxide every year. This article predictsthe adverse effects of CO۲ emissions in Iran based on time seriesdata from ۱۹۹۴ to ۲۰۲۴ for the upcoming year. In this study,statistical models such as Autoregressive Integrated MovingAverage (ARIMA), machine-learning models including linearregression and random forest, and deep learning models have beenemployed. By analyzing the performance of the models and theirpredictive capabilities, it can be observed that three models,namely SVR, ARIMA (۰,۱,۱), and Linear Regression, outperformthe other models created. The results indicate that the SVR modelis the most accurate model for predicting CO۲ emissions witherror metrics of RMSE=۶۲.۸۷, MAPE=۰.۰۶۳, and MedianAE=۴۸.۶۸, among other performance criteria. Therefore, the SVRmodel based on deep learning is suggested as one of the mostsuitable models for predicting CO۲ emissions.

Greenhouse gases, CO۲ per capita emissions,Prediction, Machine learning, Deep learning.