Predicting Corporate ESG Scores Using Machine Learning: A Comparative Study

This paper investigates the application of machine learning to predict corporate Environmental, Social, and Governance (ESG) scores, focusing on identifying the most influential factors within companies' reports. Three modelslinear regression, random forests, and gradient boostingwere utilized to estimate ESG risk scores. Hyperparameter tuning through Grid Search with cross-validation ensured that the models were optimized for robust performance, and their accuracy was assessed using Root Mean Square Error (RMSE) and R-squared (R) metrics. The findings from the experiments indicate that the gradient boosting model surpasses other methods in accuracy. According to Shapley Additive Explanations (SHAP) analysis, industry classification emerges as the primary influencer of ESG scores, followed by financial indicators such as Price/Sales, Price/Book, and Market Capitalization. This predictive model can provide valuable insights for both investors and companies, aiding in investment decisions and strategic improvements in ESG performance. While the study effectively demonstrates the potential of machine learning for ESG forecasting, future research could further refine the models by exploring more advanced ensemble methods like Extreme Gradient Boosting (XGBoost) or Categorical Boosting (CatBoost) and incorporating qualitative data to enhance predictive power.