Exploration of Dual-Carbon Target Pathways Based on Machine Learning Stacking Model and Policy Simulation—A Case Study in Northeast China

Northeast China, a traditional heavy industrial base, faces significant carbon emissions challenges. This study analyzes the drivers of carbon emissions in 35 cities from 2000–2022, utilizing a machine-learning approach based on a stacking model. A stacking model, integrating random forest and eXtreme Gradient Boosting (XGBoost) as base learners and a support vector machine (SVM) as the meta-model, outperformed individual algorithms, achieving a coefficient of determination (R2) of 0.82. Compared to traditional methods, the stacking model significantly improves prediction accuracy and stability by combining the strengths of multiple algorithms. The Shapley additive explanations (SHAP) analysis identified key drivers: total energy consumption, urbanization rate, electricity consumption, and population positively influenced emissions, while sulfur dioxide (SO2) emissions, smoke dust emissions, average temperature, and average humidity showed negative correlations. Notably, green coverage exhibited a complex, slightly positive relationship with emissions. Monte Carlo simulations of three scenarios (Baseline Scenario (BS), Aggressive De-coal Scenario (ADS), and Climate Resilience Scenario (CRS)) the projected carbon peak by 2030 under the ADS, with the lowest emissions fluctuation (standard deviation of 5) and the largest carbon emissions reduction (17.5–24.6%). The Baseline and Climate Resilience scenarios indicated a peak around 2039–2040. These findings suggest the important role of de-coalization. Targeted policy recommendations emphasize accelerating energy transition, promoting low-carbon industrial transformation, fostering green urbanization, and enhancing carbon sequestration to support Northeast China’s sustainable development and the achievement of dual-carbon goals.