Please use this identifier to cite or link to this item:
Title: Mejora en la clasificación de pacientes mediante técnicas de machine learning: aplicación a un problema neurológico a partir de la obtención de 14 biomarcadores
Author: Quintana Luque, Manuel
Director: Sánchez Pla, Alexandre
Tutor: Pérez Hoyos, Santiago
Keywords: machine learning
Issue Date: 2-Jan-2019
Publisher: Universitat Oberta de Catalunya (UOC)
Abstract: The aim of this work is to know if the use of machine learning techniques with biomarkers is useful to improve the classification of stroke patients. For this purpose, patients with suspected stroke have been obtained from a published study in which logistic regression models were used without success. Machine Learning techniques have been applied to the training data (n = 541) and the performance of the algorithms has been evaluated in a validation sample (n = 766), obtaining the diagnostic ability of the models through the use of confusion matrices. The best algorithm to classify stroke/mimic was obtained by a Random Forest trained with 10-fold crossvalidation, obtaining an accuracy of 86.7% in the validation sample. The best algorithm to classify ischemic/hemorrhagic stroke was obtained by an artificial neural network trained with 3-fold crossvalidation, with an accuracy of 86.8% obtained in the validation sample. The results of the logistic regression models have not been improved by machine learning techniques in the classification of ischemic/hemorrhagic stroke, but an improvement was achieved in the stroke/mimic classification. However, the 90% of accuracy expected at the beginning of the study has not been reached. In conclusion, the diagnostic abilities of the algorithms obtained by Machine Learning techniques are not much higher than those obtained by logistic regression. Thus, until the obtention of new potent biomarkers, it is not possible to classify these patients earlier, being still necessary the obtention of complementary medical tests.
Language: Spanish
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
maquintaTFM0119memoria.pdfMemoria del TFM1.63 MBAdobe PDFView/Open

This item is licensed under a Creative Commons License Creative Commons