Automatizando a fiscalização de gastos públicos por meio da classificação automática de empenhos utilizando aprendizado de máquina

Abstract

The budget oversight of the State of Goiás is in charge of the Court of Accounts.To accomplish this task, when the need arises, experts are assigned to manually assesseach expense. However, this process is costly given the large amount of monthly expenses.Therefore, it is necessary to automate this inspection process. One strategy for dealing with this problem is by performing sampling analysis, selecting only a subset of the expenses.Thus, many potentially wrong expenses would be unconsidered on this analysis. Another strategy is to use traditional programming techniques. However, the understandment of all data patterns is needed to create a program to deal with them. As the proposed problem is not trivial, this would be unfeasible. With this in mind, this work proposes using machine learning techniques on this inspection process, which have the ability to understand complex patterns present in the data. In result, it was possible to achieve a hit rate of 90 % according to the Micro F1 evaluation metric. In addition, this work’s secondary objective is to provide beginners in this field of study with a guide, presenting the steps to be followed and applying them to a real problem, through the data provided by the Court of Accounts of the State of Goiás.