Beyond one-size-fits-all: Differentiated green development assessment integrating a hybrid approach in China's Yangtze River Economic Belt

Abstract

Assessing green development is critical for global sustainability progress, yet existing approaches often use fixed indicators that overlook regional variations and complexities. To address this, we developed a context-specific indicator system integrated with interpretable machine learning (ML) models to balance adaptability and efficiency. Validated using 2016–2020 data from 130 Yangtze River Economic Belt cities. The results reveals notable improvements in the overall green development since 2016, with variations across city types. Random forest models exhibited high prediction accuracy (R² = 0.75–0.91, MSE = 1.91–5.77), with SHapley Additive exPlanations (SHAP) analysis highlighting rural income (|SHAP| = 1.65), tertiary industry ratio (|SHAP| = 1.48), grassland coverage (|SHAP| = 1.38), protection areas (|SHAP| = 0.81), and waste utilization (|SHAP| = 0.80) as pivotal drivers. The policy implications include inclusive development, innovation-driven industrial transformation, circular economy adoption, and ecological conservation. While temporal-spatial limitations exist, this study confirms ML's potential for complex sustainability assessment, offering decision-makers a data-driven toolkit to enhance regional intervention efficiency and effectiveness.