O uso de Machine Learning (t-SNE) como técnica para visualização de padrões no Programa Bolsa Família com base nos dados do Cadastro Único

Abstract

This work fits into a very peculiar economic context for the country, where there is great concern about the efficiency of public spending, especially on the useful use of public money for social assistance to those who really need it. In this context, this project aims to evaluate the use of machine learning in the monitoring and evaluation of benefits of the Bolsa Família Program, in order to identify abnormalities or deviations from a common standard. In order to achieve the intended objectives, research was done in government reports on this public policy, as well as in scientific articles on the t-SNE algorithm, which allows the visualization of the data through the reduction of dimensionality. With this method, it was possible to visualize social benefits in 2D and 3D graphs, which allows managers and control bodies an intelligent mechanism to generate samples used in their evaluations and inspections, as well as to focus on identifying evidence of fraud in certain benefits considered outliers. The results showed that it was possible to visualize certain groups where it is possible to act with a specific focus within the scope of the controls of the care policy analyzed. However, given the limited scope of this work, larger tests can be performed, with an increase in computational power employed and with a greater range of benefits, including new dimensions as input data.