Scholarlink

<p>The rising global emissions of CO2 affect environmental sustainability, particularly among the highest emitters, including India, Russia, Canada, and Japan. This study contributes to the literature by offering a holistic, comparative analysis of forecasting CO2 emissions using univariate time-series data spanning 1990 to 2022, with a 12-year forecast. A total of ten forecasting models were considered, including statistical (Holt-Winters, GPR, BSTS), machine learning (Decision Tree, Random Forest, Gradient Boosting, CatBoost), and deep learning (CNN-LSTM, GRU, N-BEATS), and were assessed using nine different performance metrics. India’s emissions profile has the best performance by GRU, GPR, and BSTS Forecast (DLT) in the other models. The model GRU is recognised to be highly efficient in its forecasting (in terms of the predictive accuracy) on the India emissions profile only, with a MAPE of 2.1875%, RMSE of 175.6234, and explained variance of 0.8275, global accuracy 97.91%. For the complete data of the four countries, the model GRU is also the best, to be consistent, again, with a global RMSE of 34.5049, MAPE of 2.1612%, and MedAE of 16.8876, proving its ground in the temporal dynamics of CO2 emissions. In a relative sense, GRU and GPR ranked highest among the 4 middleware for long-term forecasting of CO2 emissions, tied with CNN-LSTM and BSTS, respectively. This study has developed a consolidated methodology for estimating CO₂ emissions across multiple countries and modelling frameworks. This has important implications for the development of climate policies, emissions reduction, and sustainable energy strategies. The combination of sophisticated deep learning and probabilistic models further underscores their relevance for environmental monitoring and decision-making.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparative forecasting of CO₂ emissions in India, Russia, Canada, and Japan using statistical, machine learning and deep learning models

  • 期刊论文

  • 发表日期：

    2026年4月13日

查看全文
  • Allacheruvu Brahmaiah,
  • Kattika Koteswara Rao,
  • Kondragunta Rama Krishnaiah,
  • Nanduri Srinivas

摘要

The rising global emissions of CO2 affect environmental sustainability, particularly among the highest emitters, including India, Russia, Canada, and Japan. This study contributes to the literature by offering a holistic, comparative analysis of forecasting CO2 emissions using univariate time-series data spanning 1990 to 2022, with a 12-year forecast. A total of ten forecasting models were considered, including statistical (Holt-Winters, GPR, BSTS), machine learning (Decision Tree, Random Forest, Gradient Boosting, CatBoost), and deep learning (CNN-LSTM, GRU, N-BEATS), and were assessed using nine different performance metrics. India’s emissions profile has the best performance by GRU, GPR, and BSTS Forecast (DLT) in the other models. The model GRU is recognised to be highly efficient in its forecasting (in terms of the predictive accuracy) on the India emissions profile only, with a MAPE of 2.1875%, RMSE of 175.6234, and explained variance of 0.8275, global accuracy 97.91%. For the complete data of the four countries, the model GRU is also the best, to be consistent, again, with a global RMSE of 34.5049, MAPE of 2.1612%, and MedAE of 16.8876, proving its ground in the temporal dynamics of CO2 emissions. In a relative sense, GRU and GPR ranked highest among the 4 middleware for long-term forecasting of CO2 emissions, tied with CNN-LSTM and BSTS, respectively. This study has developed a consolidated methodology for estimating CO₂ emissions across multiple countries and modelling frameworks. This has important implications for the development of climate policies, emissions reduction, and sustainable energy strategies. The combination of sophisticated deep learning and probabilistic models further underscores their relevance for environmental monitoring and decision-making.