Uso de redes neurais para o problema de previsão de pacientes de alto custo

Detalhes bibliográficos
Ano de defesa: 2021
Autor(a) principal: Franklin Messias Barbosa
Orientador(a): Renato Porfirio Ishii
Banca de defesa: Não Informado pela instituição
Tipo de documento: Dissertação
Tipo de acesso: Acesso aberto
Idioma: por
Instituição de defesa: Fundação Universidade Federal de Mato Grosso do Sul
Programa de Pós-Graduação: Não Informado pela instituição
Departamento: Não Informado pela instituição
País: Brasil
Palavras-chave em Português:
Link de acesso: https://repositorio.ufms.br/handle/123456789/3959
Resumo: The growing aging of the world population, along with several environmental, social, and economic factors, end up posing major challenges for public health in general. Within this scenario, it is of interest for both private health insurance operators and public health managers to better manage available resources to reach the largest possible share of society. To do so, keeping in mind the amount of information produced daily, it is also clear the need for data processing and decision support technologies so that such management can be done satisfactorily. This study aims to analyze the application of machine learning and deep learning techniques in health care scenarios. One of the possible applications includes the detection of possible high-cost patients from historical data, to better target interventions that may prevent the transition of regular patients into high-cost ones or, in the case of those who are already in this condition, to allow appropriate approaches, rather than generic ones. In both cases, the detection of such patients can be beneficial, reducing avoidable costs and improving patients’ condition. The final model, chosen to predict the high-cost condition was a fully connected sequential network, with 3 hidden layers and 3 dropout layers. That network had 88% on accuracy and f1 score metrics, 91% on recall, 86% on precision and 84% specificity, showing the model’s capacity to correctly classify examples from both classes. This work also aimed to make the creation and testing of such networks easier, by providing the tools developed during its evolution on GitHub.