Environmental, social, and governance (ESG) ratings reflect a company's sustainability practices and support decision-making, yet research often overlooks industry-specific analysis, limiting cross-sector insights. This study aims to predict industry-specific ESG scores by applying ten machine learning algorithms to two U.S. industries: manufacturing and services. Results show that ESG scores are more predictable in manufacturing than in services. Of the tested algorithms, the extreme gradient boosting regressor is most effective for manufacturing, while the random forest regressor performed best in services. We employ mean decrease in impurity, permutation importance and double machine learning to identify key drivers. In manufacturing, GDP, firm size, and leverage are consistently important, with unemployment showing a strong causal effect, while service-sector results are more heterogeneous, reflecting greater complexity. This combined approach of predictive and causal analysis offers valuable industry-specific insights, advancing understanding of ESG prediction and highlighting the need for tailored strategies for different sectors.