Aprendizado de máquina para detecção de comportamento fraudulento em dados simulados do Sistema Odontológico do SUS

• Ano de defesa: 2017
• Autor(a) principal: Araújo, Ênnyo José Barros de
• Orientador(a): Gurjão, Edmar Candeia
• Banca de defesa: Carvalho, Joelson Nogueira de , Barbosa, Paulo Eduardo e Silva
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual da Paraíba
• Programa de Pós-Graduação: Programa de Pós-Graduação Profissional em Ciência e Tecnologia em Saúde - PPGCTS
• Departamento: Pró-Reitoria de Pós-Graduação e Pesquisa - PRPGP
• País: BR
• Palavras-chave em Português: Aprendizado de máquina; Detecção de fraude; Sistemas de auditoria; Sistemas de saúde
• Palavras-chave em Inglês: Machine Learning; Fraud Detection; Healhy System
• Área do conhecimento CNPq: CIENCIAS DA SAUDE
• Link de acesso: https://repositorio.uepb.edu.br/handle/123456789/72806

The Brazilian Unified Health System (Sistema Unificado de Saúde –SUS) has one of the largest free dental care programs in the world. The federal government, through its federated entities, provides dental treatment to approximately 70% of the Brazilian population. Even with such expressive numbers, this system still has many iniquities and suffers from an endemic problem in Brazil, corruption, in this case in forms of frauds. This is not a local problem, institutions around the world are studying and building technological solutions that can tackle one of the faces of frauds. Studies indicate that fraud is responsible for up to 40% of wasted resources in SUS, bringing huge financial losses, especially penalizes those who most need medical care, the low income population. Detecting and eliminating fraud is not a trivial problem, it requires technologies, experts and tools. Data mining techniques can reveal patterns difficult to be detected by humans. In this work using simulated data based on the SUS data standard machine learning is applied to detect frauds. Isolation Forest, K- Nearest Neighbors and Mahalanobis Distance algorithms was applied and sucessfully detected frauds.