Scholarlink

<p>The purposes of the study were to offer a holistic approach to the assessment of global environmental sustainability, using spatial analysis, clustering, and explainable machine learning by employing data from 180 countries of the 2024 Environmental Performance Index. The methodological approach used is based on the integration of K-means clustering, regression analysis, and explainability using the SHAP model. Spatial analysis shows that regional gaps persist with Western Europe, North America, and parts of East Asia having good levels of performance, while Sub-Saharan Africa and South Asia consistently rank the lowest. The K-Means clustering method identified five structurally different types of countries, offering differentiation beyond aggregate rankings. Finally, in the predictive phase, the best accuracy is found to be through the Stacking Regressor method (R<sup>2</sup>: 0.9997; MAE: 0.1235), followed by Gradient Boosting (R<sup>2</sup>: 0.9490) and XGBoost (R<sup>2</sup>: 0.9070). The results of the feature importance using the SHAP method have identified the top drivers to be Ecosystem Vitality (3.1394) and Environmental Health (3.0261); however, it is noticeable that Climate Change, Air Quality, Water Resources, and Sanitation/Drinking Water have also played a role. The results illustrate the value of the argument that different countries with similar overall scores may have significantly different environmental structures. The proposed framework offers a clear instrument for the development of sustainability-oriented policies.</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

Mapping latent environmental performance regimes: a spatial and explainable machine learning approach

  • 期刊论文

  • 发表日期：

    2026年4月29日

查看全文
  • Sadullah Çelik,
  • Mahmut Ünsal Şaşmaz

摘要

The purposes of the study were to offer a holistic approach to the assessment of global environmental sustainability, using spatial analysis, clustering, and explainable machine learning by employing data from 180 countries of the 2024 Environmental Performance Index. The methodological approach used is based on the integration of K-means clustering, regression analysis, and explainability using the SHAP model. Spatial analysis shows that regional gaps persist with Western Europe, North America, and parts of East Asia having good levels of performance, while Sub-Saharan Africa and South Asia consistently rank the lowest. The K-Means clustering method identified five structurally different types of countries, offering differentiation beyond aggregate rankings. Finally, in the predictive phase, the best accuracy is found to be through the Stacking Regressor method (R2: 0.9997; MAE: 0.1235), followed by Gradient Boosting (R2: 0.9490) and XGBoost (R2: 0.9070). The results of the feature importance using the SHAP method have identified the top drivers to be Ecosystem Vitality (3.1394) and Environmental Health (3.0261); however, it is noticeable that Climate Change, Air Quality, Water Resources, and Sanitation/Drinking Water have also played a role. The results illustrate the value of the argument that different countries with similar overall scores may have significantly different environmental structures. The proposed framework offers a clear instrument for the development of sustainability-oriented policies.