Ownership, regulation, and ESG in transport and logistics: Insights for policy from explainable machine learning

This study employs an explainable machine learning framework to examine the key factors associated with environmental, social, and governance (ESG) performance in China's transportation and logistics industry. Using panel data of 1912 firm-year observations from A-share listed companies (2009–2023), the CatBoost model is combined with SHapley Additive exPlanations (SHAP) for accurate and interpretable predictions. Among 11 mainstream ML algorithms, CatBoost achieves the lowest mean absolute error. Results show that management expense ratio, executive compensation, fixed asset ratio, and capital intensity positively correlated with ESG performance, while firm size, independent director ratio, and ownership concentration of top ten shareholders have negatively associated with it. SHAP reveals nonlinear relationships and complex interactions. Heterogeneity tests indicate notable differences between state-owned and private enterprises and between regulated and non-regulated industries. These findings provide evidence and guidance for stakeholders, supporting targeted ESG strategies and demonstrating the value of explainable AI in promoting sustainable development.