Aplicação de Aprendizado de Máquina e Redes Neurais Artificiais para classificação da Doença Renal Crônica através de variáveis não invasivas

• Ano de defesa: 2022
• Autor(a) principal: SILVA JUNIOR, Ronaldo dos Santos
• Orientador(a): SANTANA , Ewaldo Eder Carvalho
• Banca de defesa: SANTANA , Ewaldo Eder Carvalho , SOUSA, Nilviane Pires Silva , BARROS FILHO, Allan Kardec Duailibe , SANTOS, Giselle Cutrim de Oliveira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Maranhão
• Programa de Pós-Graduação: PROGRAMA DE PÓS-GRADUAÇÃO EM ENGENHARIA DE ELETRICIDADE/CCET
• Departamento: DEPARTAMENTO DE ENGENHARIA DA ELETRICIDADE/CCET
• País: Brasil
• Palavras-chave em Português: Aprendizado de máquina; Redes Neurais Artificiais; Doença Renal Crônica
• Palavras-chave em Inglês: Machine learning; Artificial Neural Network; Chronic Kidney Disease
• Área do conhecimento CNPq: Modelos Analíticos e de Simulação
• Link de acesso: https://tedebc.ufma.br/jspui/handle/tede/3997

Areas of Artificial Intelligence (AI), machine learning and artificial neural networks are underlined in the development of techniques, which require prior preparation of the database so that the algorithm performs the automatic classification of the database. In recent years, machine learning and statistical analysis techniques have been applied by health researchers in order to make clinical prognostic, decision-making, health care management, diagnosis and monitoring of various diseases feasible. Chronic Kidney Disease, one of the most globally prevalent NCDs, is characterized by the progressive and irreversible loss of glomerular, tubular and endocrine renal function. Early diagnosis of CKD is considered a great challenge, since in early stages the disease is characteristically asymptomatic and clinical manifestations stand out among the stages of moderate to severe renal failure. In order to diagnose CKD, Regression Logistic and ANN models were used, in which 5-fold cross validation was used in a dataset of 291 individuals over 18 years of age. The models had a good performance, both having the area under the ROC curve (AUROC) = 0.94, and the ANN obtained an accuracy and sensitivity of 87%, whereas the Regression obtained an accuracy and sensitivity of 85%. Thus, our models achieved acceptable performance for classifying CKD patients, presenting themselves as a low-cost alternative for the disease screening.