Predição de mortalidade de pacientes com traumatismo cranioencefálico no Brasil usando técnicas de aprendizagem de máquina

Abstract

This work proposes, in an original way, the use of convolutional neural networks for the prediction day mortality until the 14th day in patients with traumatic brain injury. The performance of neural networks is compared with the performance of other classic machine learning tools such as logistic regressor, multilayer perceptron, support vector machine, decision trees and random forest. In the simulation of models using neural networks, several optimization methods were used, such as RMSProp, Adam, Adamax and SGDM. The database used consists of 529 records and 16 predictor variables, and was obtained from Hospital das Clínicas (São Paulo, Brazil). Due to the presence of many missing values in the predictor variables, two procedures were proposed and evaluated for filling in the missing values, using several methods, such as decision tree, random forest, k-nearest neighbor and linear regression. The best results obtained for the prediction rate were an accuracy of 0.845 and an area under the ROC curve of 0.911