Development and performance comparison of optimized machine learning-based regression models for predicting...

Area of Science:

• Environmental Science
• Data Science
• Climate Change Modeling

Background:

• Accurate national carbon dioxide (CO2) emissions prediction is critical for energy planning and achieving sustainable, low-carbon goals.
• Effective emissions reduction strategies rely on precise forecasting of CO2 output.

Purpose of the Study:

• To develop a general machine learning model for predicting national CO2 emissions across 68 countries.
• To compare the performance of various machine learning regression models for CO2 emission forecasting.

Main Methods:

• Developed nine prediction models using Support Vector Regression, Ensemble of Trees, and Gaussian Process Regression.
• Tuned hyperparameters of these machine learning models using Bayesian optimization to enhance prediction accuracy.
• Evaluated model performance using metrics such as MSE, RMSE, MAE, MAPE, and R-squared.

Main Results:

• The optimized Gaussian Process Regression model demonstrated superior performance with an R-squared value of 0.9998.
• Key performance metrics for the best model include MSE = 106.68, RMSE = 10.328, MAE = 4.904, and MAPE = 3.38%.
• The model showed robust and accurate CO2 emission predictions across numerous countries.

Conclusions:

• The optimized Gaussian Process Regression model is a reliable and highly accurate tool for predicting national CO2 emissions.
• This predictive capability supports informed decision-making for energy strategies and climate change mitigation efforts.
• The developed model offers a promising approach for achieving a sustainable and low-carbon future through accurate emissions forecasting.