ABSTRACT
The main objective of this study is to examine the usefulness of machine learning methods for analytical procedures in the audit of financial statements. For this purpose, an ML model was developed that predicts the value of costs of social securities and other benefits based on costs of salaries and wages and detects outliers. The model was implemented with the programming language Python and its libraries and uses linear regression. The study employed a sample from the Notoria online database. It is a database containing an updated, standardized format of financial statements for all companies listed on the Warsaw Stock Exchange. The sample consisted of 5,236-year observations from the period 2003 to 2022. In the sample, there is a data of 758 firms from 11 sectors. The results indicate that the developed ML model can be useful in performing analytical procedures by auditors as it is easy in construction and interpretation, cost-effective, and has high accuracy (R-squared was equal to 91.6%). The insights are likely to be of interest to different users of financial statements. First of all, the study can be useful for audit practitioners to understand the meaning of ML and its role to improve audit efficiency and effectiveness. Second, the accountants and managers can use the study to create models to verify items of financial statements and accounting vouchers in the books. Third, the government and regulatory bodies such as Social Insurance Institutions can use the findings of the study to develop ML models targeted interventions aimed at firms that are more susceptible to avoidance of paying social contributions. The research limitations concern the fact that despite the existence of many ML techniques and algorithms, the study focuses only on one of them. The other limitation concerns the fact that this study focuses mainly on companies listed in Poland.
