| Corporations are crucial agents in practicing sustainable concepts and achieving sustainable development goals. ESG performance as standards for corporate development in Chinese?modernization, it is also an essential part of the framework ensuring the construction of a beautiful China. In the process of production and operation, corporations need to holistically improve ESG performance. Existing literature mainly focuses on the relationship between single characteristic and ESG behavior, and makes predictions only within the sample, lacking a comprehensive consideration of the potential nonlinear relationship and interactions among some of the important independent variable. According to previous literature, this paper divides the firm characteristics into four categories, namely, business strength characteristic, shareholding structure characteristic, internal governance characteristic, and external supervision pressure characteristic. Using a sample of listed firms in the Chinese A-share market from 2009 to 2020，we first examine if different dimensions of firm characteristics can predict ESG performance by using a machine learning approach: XGBoost model. The evidence shows that: 1) Compared with shareholding structure characteristic and internal governance characteristic, corporate ESG are mainly driven by business strength characteristic and external supervision pressure characteristic, which indicates that corporate ESG practice are mainly affected by business strength and external supervision pressure characteristic factors. 2) Among multiple firm characteristics, book-to-market ratio, analyst focus, corporate transparency, tangible assets ratio, net total assets ratio, cash turnover ratio, nature of ownership, external environmental governance index, management compensation and executive shareholdings have the best prediction effect on corporate ESG performance. 3) The predictive ability of machine learning method for corporate ESG performance is better than that of traditional linear research methods, and the relations between predictors and corporate ESG performance are non-linear. This paper initiates a new, more thorough perspective in effectively identifying the antecedents of corporate ESG performance using machine learning methods and has important implications for companies to better formulate and implement sustainable development strategies. |