Machine Learning Aplicado em Dados Abertos Governamentais para Detecção de Impropriedades na Aplicação de Recursos Públicos

Abstract

The high purchasing power of the State and the need to ensure the correct application of public resources raises great concern amongcitizens.In light of this reality,the legislator granted a differentiated treatment to public purchases and submitted them to greater level softransparency,determining that their data would be made available openly. In view of popular demand and the availability of access to this resource, it suse by Public Bodies that a reem poweredisimperative.However, a significant part of this data is textual, which requires diligent and specific preprocessingby those who use it.This article focuses on a data base corresponding to materials purchased by the Federal Govern mentthroughout the year 2021. The objective was to group the textual descriptions of similar purchases, allowing their comparison.Among the possible applications from the correct grouping, there is the identification of outliers in the prices of acquisitions, signalingan indication of possible impropriety.As aresult,aclustering considered satisfactory was obtained in 72% of the cases.