-
Lucas Alexandre Alvarenga Cardoso
Departamento de Automática, Universidade Federal de Lavras, MG
-
Giovanna Gouvêa Spuri de Miranda
Departamento de Automática, Universidade Federal de Lavras, MG
-
Danilo Piveta Alvarenga
Departamento de Automática, Universidade Federal de Lavras, MG
-
Arthur Phillipe Marcondes Souza
Departamento de Automática, Universidade Federal de Lavras, MG
-
Ana Cláudia Barbosa Honório Ferreira
Centro Universitário de Lavras, MG
-
Maria Helena Baena de Moraes Lopes
Escola de Enfermagem, Universidade Estadual de Campinas, SP
-
Danton Diego Ferreira
Departamento de Automática, Universidade Federal de Lavras, MG
Keywords:
confident learning, diabetes mellitus, artificial intelligence, machine learning algorithm
Abstract
Diabetic foot is a condition caused by diabetes mellitus (DM) that can cause irreversible damage, such as amputation and other complications. This article aims to classify patients diagnosed with DM in high or low risk of developing diabetic foot using the Confident Learning (CL) method. CL enables the removal of noisy data and, consequently, provides a more reliable classification. The available database is composed of 54 known risk factors of 250 patients diagnosed with DM. The development of the method was divided into: (i) feature selection; and (ii) application of the CL technique. The results showed that there is a gain of 6% in sensitivity, 14% in specificity and 9% in accuracy when removing noise from the diabetic foot ulcer risk classifier database. In addition, when comparing the competitive neural layer- based method with the proposed one, the CL-based method presented a sensitivity 26% and an accuracy 4% higher than the competitive neural layer-based method, but it has a specificity 13% lower.
