Predicting Corporate Social Ratings Using Machine Learning Models

ABSTRACT This study predicts Korean firms' environmental, social, and governance (ESG) social ratings by integrating firm‐performance indicators, human resource management indicators based on Global Reporting Initiative (GRI) Social Standards, and Job Planet‐based employee‐experience signals. Using 2945 firm‐year observations with 2020–2023 Korea Institute of Corporate Governance and Sustainability social ratings, we classify firms as “Satisfactory” or “Unsatisfactory” in a time‐aligned prediction design. Light Gradient Boosting Machine (LightGBM) is selected based on validation macro‐ F 1 after comparing machine learning and tabular deep learning models. SHapley Additive exPlanations (SHAP) identifies training investment, employment stability, welfare support, compensation satisfaction, management perception, and firm performance as key predictive signals. Prediction using 2024 data and robustness checks based on a three‐class scheme and Job Planet review thresholds support the stability of the findings. This study shows the predictive relevance of employee‐experience signals and links ESG social rating prediction with internal stakeholder management and explainable machine learning.