Predicting Carbon Emissions with Electricity Data Based on Three Algorithms: Decision Tree, Random Forest, and Extreme Gradient Boosting

The power sector, a significant contributor to carbon emissions, accounts for about <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{40 \%}$</tex> of total emissions in society. Accurate and timely carbon monitoring technology is crucial for supporting the industry's low-carbon transition. Current monitoring systems mainly rely on the actual measurement method or accounting method, often facing challenges in optimizing the detection cost and timeliness. This paper mainly primarily employs the three algorithms in machine learning: Decision Tree, Random Forest, and C. Extreme Gradient Boosting (XGBoost), aiming to leverage insights for carbon reduction through electricity use, ultimately decreasing electricity costs. The K-fold cross-validation technique is adopted in the model training to help construct a new type of real-time monitoring system for carbon emissions. Among the three algorithms, the Random Forest algorithm has a smaller gap between predicted and real values, exhibiting greater stability and generalization capability. In contrast, Decision trees and XGBoost fit less well than random forests, but the overall fit is still good.