Scholarlink

<p>The study creates a thorough machine learning solution for carbon price prediction that enables environmental policy design, together with emissions control and benefits financial operations. This research analyzes six modeling approaches consisting of Linear Regression, Ridge Regression, Elastic Net, Random Forest, XGBoost, and CatBoost. Procedures of imputation followed by standardization enabled the preparation of historical carbon price data before partitioning them into training and testing subsets. The schemes’ evaluation employs mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), coefficient of determination (R²), variance accounted for (VAF), and Explained Variance (EV), supported by visualization tools such as scatter, residual, and feature importance plots. Random Forest and XGBoost deliver the highest training performance scores, including RMSEs measuring 52004.30 and 85016.72, together with R² scores at 0.980 and 0.947, respectively. These predictive models exhibit overfitting behavior because they deliver reduced performance throughout testing. Ridge Regression proves to be the optimal choice among the other models because it provides the best combination of robustness and predictability with test MSE of 1.71 × 10¹⁰, RMSE of 130875.7, R² of 0.9488, and VAF of 0.948. The result highlights the importance of model selection and tuning for unstable and highly nonlinear data, such as carbon prices. This framework provides actionable guidelines for carbon trading elements and sustainable predictive models that industry professionals, researchers, and policymakers can use for better scheme development and planning forecasts.</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

Carbon Price Forecasting: Utilizing Historical Data for Effective Emissions Reduction and Environmental Policy Planning

  • 期刊论文

  • 发表日期：

    2025年12月20日

查看全文
  • Xuefeng liu,
  • Yuehua Bai,
  • Li Ma

摘要

The study creates a thorough machine learning solution for carbon price prediction that enables environmental policy design, together with emissions control and benefits financial operations. This research analyzes six modeling approaches consisting of Linear Regression, Ridge Regression, Elastic Net, Random Forest, XGBoost, and CatBoost. Procedures of imputation followed by standardization enabled the preparation of historical carbon price data before partitioning them into training and testing subsets. The schemes’ evaluation employs mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), coefficient of determination (R²), variance accounted for (VAF), and Explained Variance (EV), supported by visualization tools such as scatter, residual, and feature importance plots. Random Forest and XGBoost deliver the highest training performance scores, including RMSEs measuring 52004.30 and 85016.72, together with R² scores at 0.980 and 0.947, respectively. These predictive models exhibit overfitting behavior because they deliver reduced performance throughout testing. Ridge Regression proves to be the optimal choice among the other models because it provides the best combination of robustness and predictability with test MSE of 1.71 × 10¹⁰, RMSE of 130875.7, R² of 0.9488, and VAF of 0.948. The result highlights the importance of model selection and tuning for unstable and highly nonlinear data, such as carbon prices. This framework provides actionable guidelines for carbon trading elements and sustainable predictive models that industry professionals, researchers, and policymakers can use for better scheme development and planning forecasts.