Análise exploratória e comparativa da aplicação de agrupamento para combate à lavagem de dinheiro

• Ano de defesa: 2019
• Autor(a) principal: Nunes, Fabio Mangueira da Cruz
• Orientador(a): Rodrigues Júnior, Methanias Colaço
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Programa de Pós-Graduação: Pós-Graduação em Ciência da Computação
• Palavras-chave em Português: Computação; Lavagem de dinheiro; Mineração de dados (Computação); Algoritmos computacionais; Money laundering; Data mining; Data analytics; K-Means; Expectation–Maximization (EM)
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://ri.ufs.br/jspui/handle/riufs/11607

Context: Since 2007, through the National Anti-Corruption and Money Laundering Strategy (ENCCLA), the first LABLDs have been created, which are present now in all the regions of the federation and are responsible for policies to develop methods and advanced technologies to support the bodies of criminal prosecution. The need for innovation in this crime combat scene imposes partnerships, support, research and scientific method. The objective of this work was to evaluate the effectiveness of the Expectation-Maximization (EM) and K-Means algorithms on real financial transaction databases investigated by Sergipe's LABLDs, comparing the evidences found with the results obtained by mapping the state-of-the-art published in the literature. Method: Initially, a Survey was conducted with the premise of characterizing the use of techniques of storage, integration, Data Mining and Data Analytics by LABLDs and other investigative units throughout Brazil. Then, a systematic mapping was performed as a way to identify and systematize the main approaches, techniques and algorithms used in computer science to combat LD. Finally, a controlled in vivo experiment was designed and executed to compare the EM and K-Means algorithms. Results: It was found that approximately 97% of survey respondents did not directly use any data mining algorithm and that 30.99% evaluated their own knowledge about the subject as bad or very bad. Related to the state of the art, it has been identified that the main approaches used against LD are supervised classifiers and clusters. With the execution of the experimental process, it was evidenced that the algorithm EM surpasses the K-means algorithm, reaching a maximum average accuracy of 98.25%. Conclusions: This thesis exposed a hard reality within the main investigation and control bodies of our country. After analyzing the state of the art, it was evidenced that there are opportunities to explore solutions against LD, especially in the areas of Machine Learning and Deep Learning. Finally, the EM algorithm presented as a superior alternative to K-means for the implementation of a predictor module of suspicious transactions, confirming the results of the literature, however, in a real and specific investigation environment.