A Machine Learning Approach to Identifying Suspicious Tax Evasion Behavior in Public Financial Data

Abstract

The protection of sensitive tax data poses a challenge for data processors seeking to use it freely, particularly in the realm of financial and tax data. Detecting suspicious behavior related to tax evasion in this area requires sophisticated techniques. This paper presents a study on detecting tax evasion behavior using publicly available financial data. The data was obtained from a public source, and machine learning was applied to identify suspicious behavior based on criteria derived from a literature review. Our study yielded a precision of over 97 % using selected supervised models. Future work involves exploring behavior using unsupervised methods, increasing the number of instances using data synthesis techniques, and comparing results with enterprises that are not registered in a stock market as long as the data is legally available. This research provides valuable insights into identifying suspicious tax evasion behavior and can inform the development of improved tax evasion detection systems in the future.