Scholarlink

<p>Identifying key drivers of low-carbon development (LCD) in the Yangtze River Delta is essential for advancing China’s sustainable development agenda. This study introduces an interpretable machine learning framework based on the SMOTE-XGBoost model, which achieves high predictive accuracy (AUC = 0.988) in evaluating regional LCD. SHapley Additive Explanations and partial dependence plots are employed to quantify the effects of input variables. Fiscal revenue, ventilation coefficient (VC), and agricultural value added emerge as the most influential determinants. Moving beyond linear assumptions, the analysis reveals complex non-linear interactions. Specifically, all three features follow a similar pattern whereby, upon reaching a feature-specific threshold, their marginal effects on regional LCD plateau. In addition, we identify an interactive mechanism under which LCD is most sensitive to fiscal revenue in settings with limited ventilation (VC ≤ 1223&#xa0;m²/s). As fiscal capacity strengthens (fiscal revenue &gt; 2.3&#xa0;billion yuan), LCD becomes relatively more sensitive to agricultural value added, warranting a reallocation of resources toward agricultural mitigation. This shift is particularly salient in low-ventilation counties (VC &lt; 1338&#xa0;m²/s). Overall, the findings provide empirical evidence of complex, non-additive relationships among socioeconomic and environmental factors and offer actionable, data-driven insights for designing synergistic, targeted policies to support high-quality, sustainable regional development.</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

Quantifying drivers of low-carbon development in Yangtze river delta: interpretability analysis from machine learning perspective

  • 期刊论文

  • 发表日期：

    2026年2月21日

查看全文
  • Yali Hou,
  • Qunwei Wang,
  • Rong Wang,
  • Tao Tan

摘要

Identifying key drivers of low-carbon development (LCD) in the Yangtze River Delta is essential for advancing China’s sustainable development agenda. This study introduces an interpretable machine learning framework based on the SMOTE-XGBoost model, which achieves high predictive accuracy (AUC = 0.988) in evaluating regional LCD. SHapley Additive Explanations and partial dependence plots are employed to quantify the effects of input variables. Fiscal revenue, ventilation coefficient (VC), and agricultural value added emerge as the most influential determinants. Moving beyond linear assumptions, the analysis reveals complex non-linear interactions. Specifically, all three features follow a similar pattern whereby, upon reaching a feature-specific threshold, their marginal effects on regional LCD plateau. In addition, we identify an interactive mechanism under which LCD is most sensitive to fiscal revenue in settings with limited ventilation (VC ≤ 1223 m²/s). As fiscal capacity strengthens (fiscal revenue > 2.3 billion yuan), LCD becomes relatively more sensitive to agricultural value added, warranting a reallocation of resources toward agricultural mitigation. This shift is particularly salient in low-ventilation counties (VC < 1338 m²/s). Overall, the findings provide empirical evidence of complex, non-additive relationships among socioeconomic and environmental factors and offer actionable, data-driven insights for designing synergistic, targeted policies to support high-quality, sustainable regional development.