Interpretable predictive model for listed companies ESG greenwashing based on XGBoost and SHAP.

The prediction of ESG greenwashing among listed companies is crucial for ex-ante control of deceptive ESG information disclosures by companies and for early warning of ESG investment risks. Machine learning techniques are commonly used to predict corporate behavior, but their application in ESG greenwashing is limited and lacks interpretability. This study proposes a predictive model for ESG greenwashing in listed companies using the enhanced XGBoost algorithm and SHAP interpretation method. Using a dataset of Chinese listed companies from 2009 to 2022, this study selects 16 indicators of corporate and external pressure characteristics as the model's input variables, combines five-fold cross-validation and grid search parameter tuning, and constructs two company ESG greenwashing prediction models. The prediction performance is compared with Random Forest, SVM, LightGBM, BP neural network, and three XGBoost methods. The SHAP method is used to explain the contribution of the main indicators in predicting the company's ESG greenwashing. The results show that the one-period-lagged corporate characteristics model achieves 86.82% prediction accuracy, and corporate financial characteristics have a greater impact on prediction results than corporate governance characteristics. Further analysis shows that the one-period-lagged corporate characteristics model performs better in non-heavy-pollution industries and state-owned enterprises.