Predição e identificação de empresas noteiras utilizando machine learning na Secretaria de Fazenda do Ceará

Abstract

Noteiras companies are created with the aim of issuing fraudulent tax documents, which do not correspond to an effective circulation of goods or actual provision of services, with the aim of generating undue ICMS credits to be used by the recipients of the tax documents., who will be able to use these “bad” credits to offset the value of ICMS owed to the state tax authorities. In this sense, the present work consists of a review of the main experiences of using Machine Learning models for identifying and forecasting noteiras companies in Tax Administrations, as well as the main characteristics of ICMS and the taxation systems adopted. The work also carries out a theoretical study of the main elements present in a noteiras company, “orange” and “front” partners, and classifies these noteiras companies into three types according to the degree of complexity. Next, a study is carried out with five Machine Learning models for classification (Logistic Regression, KNN, Neural Network, Random Forest and XGBoost), aiming to identify and predict noteiras companies in the Ceará State Finance Department. Finally, a comparison is made between the model evaluation metrics to define which models obtained the best results.