Modelos para detecção de fraudes utilizando técnicas de aprendizado de máquina

Content Provider	Semantic Scholar
Author	Junior, P. Schiavinatto Carlos, João Sequeira
Copyright Year	2018
Abstract	Due to the massification of the credit concession in Brazil, mainly caused by recent technological development, fraud mitigation has become essential in financial institutions. Even with nowadays low occurrence rates, frauds have shown a significant increasing tendency for the future, causing, in this way, a negative impact on the organizations results. In this context, investments in more sophisticated techniques for detecting fraud has happen frequently, and in many cases, methods using Machine Learning techniques are been applied, in order to obtain more accurate and reliable predictions against fraud events. As a result of these context, this work aims to propose models and techniques that use Machine Learning in a real database, in order to compare the results obtained with traditional techniques that apply Logistic Regression techniques. Additionally, the challenge of this work was to propose a Random Forest classifier capable of identifying 3 distinct events at the same time, which may constitute a fraud. The results evidenced the viability of using a single model, as opposed to current techniques that employing multiple models, e.g. one model for each event, with a low loss of performance that can be compensated by the reduced complexity in the model implantation.
File Format	PDF HTM / HTML
Alternate Webpage(s)	http://bibliotecadigital.fgv.br/dspace/bitstream/handle/10438/27166/Dissertacao_Joao_Carlos_Pacheco_VFinal_2.pdf?isAllowed=y&sequence=3
Language	English
Access Restriction	Open
Content Type	Text
Resource Type	Article