The use of feature selection methods in the materiality analysis of banks’ ESG data

Banks play a key role in financing sustainable development and managing ESG risks, which imposes an obligation on them to adopt best practices and standards related to ESG. The main research objective is to present and evaluate the usefulness of selected feature selection methods in analysing the materiality of banks’ ESG data. To investigate the research problem, we used a sample of all European banks available in the Refinitiv (currently LSEG) database for the period 2018-2022. Feature selection and feature importance evaluation techniques, including Forward Sequential Feature Selection, Mutual Information, F-statistics, and Random Forest feature importance scores, were utilized to identify the most influential ESG indicators on the bank’s final ESG score. The use of feature selection methods allowed for the identification of key ESG indicators for banks, some of which differed from the standard indicators used in ESRS, SFRD, and SASB. In the environmental domain, the methods did not recognize CO 2 emissions as significant, while in the social and corporate governance areas, they highlighted the importance of features not included in existing standards. The results indicate that machine learning-based feature selection methods can support the analysis of the materiality of banks’ ESG data and their regulatory compliance.