Find who is doing social good: using machine learning to predict corporate social responsibility performance

Through a machine learning approach, this study develops a determinant model of corporate social responsibility (CSR) performance and comprehensively examines the predictiveness of chief executive officer (CEO) characteristics, board characteristics, firm characteristics, and industry characteristics. The results show that the extreme gradient boosting (XGBoost) model predicts CSR performance in the Chinese context more accurately than the other machine learning models tested. Moreover, the interpretable model based on the XGBoost and Shapley additive explanations (SHAP) method suggests that return on assets (ROA) has the strongest predictive power for CSR performance compared to other feature variables, followed by industry competition, firm size, industry size, customer concentration, leverage, industry growth, CEO pay, ownership, and CEO shares. Specifically, ROA, industry competition, firm size, industry size, industry growth, CEO pay, and ownership positively relate to CSR performance. In contrast, the effects of customer concentration, leverage, CEO shares, sales growth, and board diversity are negative. Overall, our study adds knowledge to sustainable operations management literature by providing insights into the use of advanced machine learning methods to predict CSR performance in the context of emerging markets, thereby offering significant implications for managers, investors, policymakers, and regulators.