Identificação de possíveis redes de conluio em licitações públicas utilizando teoria dos grafos, clusterização e PSO

• Autor(a) principal: Couto, Cecília de Freitas Vieira
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UFPB
• Texto Completo: https://repositorio.ufpb.br/jspui/handle/123456789/29939
• topic: Licitações públicas; Fraudes em licitações; K-means; Detecção de conluios; Public procurements; Collusion; Graphs; K-means; PSO; CNPQ::ENGENHARIAS::ENGENHARIA MECANICA

Public tenders are a means of contracting which seeks to ensure real competition between participants, thus avoiding the occurrence of irregularities. Despite this, it is not uncommon to identify numerous types of fraud in these bidding processes, among which the formation of collusion stands out, which occurs when two or more companies come together to defraud a bid. Despite being a practice committed for many years, the detection of collusions presents many difficulties, especially due to the lack of tools and techniques to assist the investigation process. Over the years, many methods have been developed to help this process, but due to technical limitations, none of them has been definitively established. Therefore, in view of this, this work aimed to develop a simple method, but with high power of use, to help identify the occurrence of fraud in public tenders. Graph theory techniques and the k-means algorithm optimized by PSO were used to identify suspicious relationships between companies. The proposed methodology was applied to data from public tenders that took place in the state of Paraíba between the years 2014 and 2021, made available by the Court of Auditors of the State of Paraíba (TCE/PB) through the Sagres Online portal. Thus, initially the dataset obtained underwent pre-processing and a correction of inconsistencies. Then, graphical visualizations were elaborated using graphs to exemplify the use of the tool. After that, the worked data were grouped and groups of companies with similar behavior patterns were identified. Finally, the results obtained with the application of graph theory were united with the clustering products to obtain the final result of this dissertation, which showed the great potential of the proposed method.