Corporate ESG rating prediction based on XGBoost-SHAP interpretable machine learning model

The prediction of corporate ESG ratings is of paramount importance in augmenting the scientific rigor and precision of ESG investment decisions and steering corporate management of ESG-related risks. While machine learning methodologies have been extensively utilized in forecasting corporate behavio...

Full description

Saved in:
Bibliographic Details
Published in:Expert systems with applications Vol. 295; p. 128809
Main Authors: Zhang, Jianfeng, Zhao, Zexin
Format: Journal Article
Language:English
Published: Elsevier Ltd 01.01.2026
Subjects:
ESG rating prediction
Machine learning
SHAP
XGBoost algorithm
XGBoost algorithm
SHAP
ESG rating prediction
Machine learning
ISSN:0957-4174
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The prediction of corporate ESG ratings is of paramount importance in augmenting the scientific rigor and precision of ESG investment decisions and steering corporate management of ESG-related risks. While machine learning methodologies have been extensively utilized in forecasting corporate behavior, their deployment in corporate ESG ratings remains relatively nascent and is often criticized for a lack of interpretability. This study develops a predictive model for corporate ESG ratings using an XGBoost algorithm enhanced with SHAP interpretability. The methodological framework incorporates SMOTE-ENN for handling class imbalance and a comprehensive optimization approach utilizing 3-fold cross-validation and randomized hyperparameter search. The model incorporates a comprehensive set of 15 indicators spanning four critical dimensions—financial performance, environmental impact, social responsibility, and corporate governance, using a dataset of Chinese A-share listed companies from 2013 to 2022. The model’s predictive efficacy is subsequently elucidated, revealing that the XGBoost-SHAP framework achieves an accuracy and precision rate of 91.0% and 90.7%, respectively, with an F1-score and AUC value of 90.1% and 0.977, outperforming comparative models. The analysis underscores the significant influence of financial and non-financial factors on ESG rating predictions, with financial attributes exerting a relatively more pronounced impact than individual non-financial metrics. Tailored to the diverse objectives of ESG investors, this research further delineates a level definition model and a risk identification model, achieving predictive accuracies of 92.4% and 97.7%, respectively. The insights from this study furnish ESG investors with a robust foundation for enhancing investment outcomes and offer strategic guidance for corporations aiming to elevate their ESG performance.
ISSN:0957-4174
DOI:10.1016/j.eswa.2025.128809