Banca de DEFESA: Rodrigo Vilela Fonseca de Souza

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : Rodrigo Vilela Fonseca de Souza
DATE: 20/06/2023
TIME: 16:00
LOCAL: https://teams.microsoft.com/l/meetup-join/19%3a0a9998084fcc459f92b72fb6608767f7%40thread.tacv2/16865
TITLE:

Collusion identification in Comprasnet auctions with Machine Learning.

KEY WORDS:

Auction, Collusion, Machine Learning, Cartel.

PAGES: 71
BIG AREA: Engenharias
AREA: Engenharia Elétrica
SUMMARY:

The Brazilian Federal Government executes a large volume of public procurements through the Comprasnet Procurement Portal, which is a website for electronic auctions available for bidders nationwide and abroad. In the period from 2018 to 2021, approximately R\$144 billion bids were applied within Comprasnet, with the execution of more than 122 thousand processes of this modality. The audit of these events is one of the duties of the Federal Comptroller General (Controladoria Geral da União - CGU) agency, which has developed tools to support such audit activities, especially involving a large volume of data processing. Thus, it is possible for electronic trading sessions to be audited in time to identify irregularities and rectify them. Between 2019 e 2020, following CGU preventive actions, around R$ 6.7 billion auctions were revoked, suspended or adjusted. Among the irregularities, collusion is difficult to identify, given the set of variables involved in the process. Artificial Intelligence applied to data analysis, through Machine Learning algorithms, presents itself as a promising method towards the detection of collusion between the auction's participants. In this work, a study of machine learning algorithms was carried out, in 4 different scenarios, on two datasets extracted from Comprasnet and other published collusion datasets. In the best scenarios, ensemble methods algorithms achieved an accuracy greater than 87%. Considering all metrics applied, Extra Trees was the algorithm with the best performance, capable of indicating new possible collusion cases.

BANKING MEMBERS:
Externo à Instituição - FABIANO CAVALCANTI FERNANDES - IFB
Interno - 906.575.601-97 - DANIEL ALVES DA SILVA - UnB
Presidente - 2311780 - FABIO LUCIO LOPES DE MENDONCA
Interno - 2556078 - GEORGES DANIEL AMVAME NZE