Ciência de Dados aplicada na detecção de fraude em demonstrações financeiras

Abstract

This study aims to understand the application of Data Science and Benford's law as auxiliary instruments in the auditing of financial transactions. As a case study, the conformity of extra-budgetary disbursements of the State of Goiás for the fiscal year 2024, available on the State of Goiás Transparency Portal, is analyzed. The objective is to verify if the transactions are free from distortions indicating possible irregularities, through the application of Benford's Law supported by statistical tests. The methodology adopts the Knowledge Discovery in Databases (KDD) process, covering data collection, cleaning, mining, and validation using Chi-Square and Z-score statistical tests, implemented in Python. The results demonstrate significant divergences between the observed frequency and that expected by Benford's Law, highlighting a higher-than-expected occurrence of the first digit 1 and compensatory deviations in other digits, leading to the rejection of the null hypothesis of conformity. It is concluded that the developed computational tool acts effectively as an initial filter for detecting anomalous patterns in audits, allowing the auditor to focus on higher-risk transactions and optimize inspection resources.