Um módulo inteligente baseado em aprendizado de máquina para treinamento de estudantes de medicina no doctraining

• Autor(a) principal: Silva, Jesaías Carvalho Pereira
• Data de Publicação: 2020
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Digital da Universidade Federal Rural do Semi-Árido (RDU)
• Texto Completo: https://repositorio.ufersa.edu.br/handle/prefix/5189
• topic: Aprendizado de Máquina; Medicina; Jogo; Múltiplos Classificadores; Casos clínicos; Machine Learning; Medicine; Game; Ensemble Classifiers; Clinical Cases; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO

In recent years, medical students from Brazil have obtained poor performance in the Regional Medical Council of São Paulo State exam. Although the most recent evaluation shows improvements, they still perform poorly in specialties such as clinical cases, diabetes and myocardial infarction. In this context, several strategies emerge to improve the teaching and learning of medical students, one of which is serious games. However, there is a problem when trying to train medical students in clinical cases through a game, as there are a large number of diseases, and processing a database containing these diseases is laborious, and it is always necessary to be with this updated information, as over time new diseases arise and symptoms may vary. It is necessary to update the data so that students have access to the most recent. This research proposes an intelligent module to assist in the management of a serious game, performing the classification of medical data through machine learning models and making them available to a serious game. In addition, it is also necessary that these data are easily administered by medical professionals who use the smart module. This smart module is part of DocTraining, a project that has a serious game for training medical students. This module has a website for disseminating, managing and expanding data. The information is made available for the serious game through a web service. Databases of clinical cases, heart disease, and diabetes were used. These databases are labeled using multiple machine learning classifiers that aggregate the classifications through trust and majority vote. The intelligent module was validated computationally and with medical professionals. In computational validation, the classifiers used the metric of Precision and F1 Score, using percentage split and cross-validation. In relation to validation with professionals, we conduct a semi-structured interview a professor and a laboratory technician from the medical course at the Federal Rural University of the Semi-Arid. With the results found, it was possible to observe that the combination of classifiers improved the initial results. Highlighting the combination of Naive Bayes, Support vector machine and Logistic regression using Confidence to classify the basis of heart disease; Support vector machine, Decision Tree and Logistic Regression using Majority Vote for classification of Diabetes; and Support vector machine, K-Nearest Neighbor, Naive Bayes and Logistic regression using Confidence for classification based on Clinical Cases. The professionals have shown that DocTraining intelligent module can manage serious game data, with the potential to evolve and become a tool capable of assisting the teaching-learning of medical students and teachers