Um modelo estratégico para a análise de crédito utilizando redes neurais artificiais

• Ano de defesa: 2008
• Autor(a) principal: Pires, Íris Rosane Netto
• Orientador(a): Yamanaka, Keiji
• Banca de defesa: Lima, Luciano Vieira , Soares, Alexsandro Santos
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia
• Programa de Pós-Graduação: Programa de Pós-graduação em Engenharia Elétrica
• Departamento: Engenharias
• País: BR
• Palavras-chave em Português: Crédito; Análise de crédito; Análise subjetiva de crédito; Ativos intangíveis; Redes neurais artificiais; Administração de crédito; Redes neurais (Computação)
• Palavras-chave em Inglês: Credit; Credit analysis; Subjective credit analysis; Intangible assets; Artificial neural networks
• Área do conhecimento CNPq: CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: https://repositorio.ufu.br/handle/123456789/14390

The present work reflects a case study done at a factoring company, acting in the area of market fomenting. This work focuses on the negotiations related to financial credit done with its client businesses, notably micro and small companies. It examines the credit analysis system utilized, highlighting that, after a selected bibliographical review and a deep study of its method and analysis process, some gaps were found which give its negotiations credit concession actions of high risk. With the objective of giving support to the analysts of the aforementioned company in the analysis process, as well as conferring a more efficient directive for decision making, this work proposes a strategic credit analysis model which uses contemporary approaches with the objective of aggregating value and giving new emphasis to the model used. The proposed model is based on the utilization of two conceptual tools: subjective credit analysis, applied in a standardized way and concepts of intangible assets to qualify and quantify the portfolio management risks; and a computational tool, artificial neural network techniques specifically, to process, learn and generalize the proposed model, and from there, form a more accurate diagnosis for future clients. Therefore, after the elaboration of the method, several tests were made, the results of which were considered promising and with a good proficiency level. Not only was the contemporary approach utilized considered prominent, but also the application of neural networks demonstrated high performance handling the multivariate data given to it.