ABSTRACT
The chapter examines the usefulness of explainable artificial intelligence (XAI) in the attestation of sustainability reporting. For this purpose, a machine learning (ML) model was developed to predict one of the sustainability metrics: the gender diversity of key executives based on a sample of 718 public Polish companies from the EMIS online database. The complex model Random Forest with bootstrap aggregating was implemented using Python and its libraries. Despite its high accuracy of 92.3%, the model does not explain the impact of input variables. The SHapley Additive explanation (SHAP), one of the XAI techniques, was employed to describe the global and local contributions using the feature importance. The findings show that the number of key executives had the most significant impact on the model output, followed by the company’s size measured by the natural logarithm of capitalization, assets, and revenue. Furthermore, companies with five or fewer key executives tend to exhibit twice the average level of gender diversity compared to firms with a greater number of board members and proxies. The results confirm that ML algorithms and XAI tools such as SHAP can be beneficial for auditors in performing sustainability analytics to find hidden patterns and relationships. The insights will likely interest different users of sustainability reporting, managers, and auditors.
