Please use this identifier to cite or link to this item:
Title: Exploring dimensionality reduction and machine learning methods for the prediction of body composition abnormalities among an HIV+ population
Author: Pelegrín Cuartero, Carolina
Director: Ventura Royo, Carles  
Tutor: Pérez Álvarez, Nuria
Keywords: dimensionality reduction
machine learning
body composition
Issue Date: 5-Jun-2019
Publisher: Universitat Oberta de Catalunya (UOC)
Abstract: The main aim of this study was to classify a set of patients with HIV as having different type of body abnormalities (i.e. osteoporosis/osteopenia, lipodystrophy, low muscle mass), caused by the antiretroviral therapy and the chronic inflammation of the immune system caused by the virus itself. Building classifiers may lead to earlier diagnose, decreasing health effects and improving life quality and expectancy of HIV+ patients. For this study, a set of measurements from a DEXA analysis was available; three of them were used to establish the presence of disease, based on cut-offs found at the bibliography. Models were built with original ("raw") variables and synthetic variables created by principal component analysis, multiple factor analysis and clustering of variables. For the prediction of each disease, just not-directly-related features were taken into account. Different type of classification methods were used, including logistic regression, machine learning and ensemble learning methods. Models were fitted using training datasets and validated using test datasets; "best" models were selected based upon their accuracy and AUC value. Ensemble models greatly improved prediction of lipodystrophy and low muscle mass, with models showing an excellent performance, demonstrating its capacity to extract subtle patterns from the data. Performance of models for the prediction of bone-related disease was just acceptable, probably due to the class-imbalance present and the lack of important variables related to the bone quality.
Language: English
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
cpelegrincTFM0619memory.pdfMemory of TFM1.43 MBAdobe PDFView/Open

This item is licensed under a Creative Commons License Creative Commons