Tax evasion identification using open data and artificial intelligence

Abstract

Tax evasion is the practice of the non-payment of taxes. In Brazil alone, it is estimated as 8% of GDP. Thus, governments must use intelligent systems to support tax auditors to identify tax evaders. Such systems seek to recognize patterns and rely on sensitive taxpayer data that is protected by law and difficult to access. This research presents a smart solution, capable of identifying the profile of potential tax evaders, using only open and public data, made available by the Brazilian internal revenue service, the administrative council of tax appeals of the State of Goiás, and other public sources. Three models were generated using Random Forest, Neural Networks, and Graphs. The validation after fine improvements offered an accuracy greater than 98% in predicting tax evading companies. Finally, a web-based solution was created to be used and validated by tax auditors of the State of Goiás.

Keywords:
tax evasion; neural networks; artificial intelligence; open data; tax auditing