Machine learning algorithms for predicting ESG scores: evidence from the Stock Exchange of Thailand

This research applies machine learning algorithms to develop models for predicting sustainability ratings, focusing on both the overall Environmental, Social, and Governance (ESG) score and the individual pillar scores: environment (E score), social (S score), and governance (G score). These scores are essential for investors, regulators, and companies assessing corporate sustainability performance. The models evaluated include linear regression, decision trees, random forests, and artificial neural networks, using financial data from 2020 to 2022. Prediction accuracy was measured using the coefficient of determination (R²), mean absolute error (MAE), root mean squared error (RMSE), and mean absolute percentage error (MAPE). Among the models, the random forest model demonstrated superior performance, achieving the highest R² value of 0.47 for the overall ESG score, along with lower error rates (MAE: 0.09, RMSE: 0.12, and MAPE: 23.52%). The results also indicate that company size significantly influences the prediction of ESG scores, particularly in the environmental and social dimensions. These findings highlight the random forest model’s robust predictive capability, making it a reliable tool for ESG score assessment. This study demonstrates the potential of machine learning to improve sustainability ratings.