APPLICATION OF BENFORD’S LAW IN THE ANALYSIS OF THE RELIABILITY OF FINANCIAL STATEMENTS USING MACHINE LEARNING METHODS

The purpose of the study is to build an algorithm for assessing the reliability of financial statements based on the integration of digital patterns identified by Benford's law with anomaly analysis procedures based on machine learning. The research materials include empirical models of the distribution of numbers, methods of primary filtering of financial data, as well as tools for detecting deviations based on isolation trees, support vectors and probabilistic estimates. The paper uses logical and structural analysis, conceptual design, and comparative formalization of data discovery procedures. Research result: a hybrid algorithm has been developed that combines digit verification with feature vectorization and subsequent aggregation of solutions from independent anomaly detection models. Conclusions: a) the isolated application of Benford's law does not provide the necessary accuracy of analysis when the structure of the reporting data is complex; b) the algorithmic combination of digital and statistical features makes it possible to form reasonable estimates of the degree of deviation; c) the proposed scheme is suitable for use in audit monitoring and regulatory analysis procedures.