An optimized machine learning framework for predicting and interpreting corporate ESG greenwashing behavior.

Area of Science:

• Environmental, Social, and Governance (ESG) analysis
• Machine Learning applications in finance
• Corporate sustainability and transparency

Background:

• Accurate prediction and interpretation of corporate ESG greenwashing are vital for transparency and regulatory effectiveness.
• Existing prediction models face limitations in hyperparameter optimization and interpretability.
• Need for advanced frameworks to address these challenges in ESG disclosure.

Purpose of the Study:

• To develop and validate an optimized machine learning framework for predicting and interpreting corporate ESG greenwashing behavior.
• To enhance the accuracy and interpretability of ESG greenwashing detection models.
• To provide actionable insights for regulators and investors.

Main Methods:

• Developed a comprehensive ESG greenwashing prediction dataset.
• Integrated Improved Hunter-Prey Optimization (IHPO) for hyperparameter tuning of eXtreme Gradient Boosting (XGBoost) models.
• Utilized SHapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• The IHPO-XGBoost model demonstrated superior performance in predicting ESG greenwashing (R²=0.9790, RMSE=0.1376, MAE=0.1000).
• Achieved higher accuracy compared to traditional HPO-XGBoost and other optimization algorithms.
• SHAP analysis identified key features and their interactions influencing prediction outcomes.

Conclusions:

• The IHPO-XGBoost framework offers a robust solution for predicting and interpreting corporate ESG greenwashing.
• Enhanced interpretability provides crucial insights into factors driving greenwashing behavior.
• Findings support improved regulatory efficiency and informed investment decisions in ESG contexts.