Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.728653
Title: The waterfall support vector machine for the multiclass imbalance problem : an application to predict retinopathy of prematurity in neo-natal infants
Author: Rollins, Rebecca
Awarding Body: Queen's University Belfast
Current Institution: Queen's University Belfast
Date of Award: 2017
Availability of Full Text:
Full text unavailable from EThOS.
Please contact the current institution’s library for further details.
Abstract:
Retinopathy of prematurity (ROP) is a disease in which retinal blood vessels of premature infants fail to develop properly. While it is possible that the disease can regress itself, it can also result in visual impairment and even blindness. The incidence of ROP in Northern Ireland is increasing, as is the pressure on neonatal wards and hospital staff. One way to help would be to identify those infants at risk, and prioritise screenings accordingly, while potentially reducing the number of unnecessary screenings for low risk infants. This thesis adds to the current literature on Discrete Conditional Survival models (DC-S). A model which consists of two components; a conditional component and a process component. The main aim of this research is to develop a model which can accurately predict multiple patient outcomes of a rare disease. This was achieved by extending a support vector machine (SVM) to work for multiple classes, where the class distribution was imbalanced. The new Waterfall SVM with SMOTE is considered as an alternative conditional component of the DC-S model. The Waterfall SVM is applied to neonatal data which shows how it can be used to identify infants at risk of ROP. The model performed very well but to test its ability to generalise, it was applied to five datasets and compared with other approaches. For both the F-measure and G-mean the Waterfall SVM with SMOTE gave the best performance for four out of the six tasks, and performed admirably in the others. The length of stay in neonatal care for infants from each of the respective outcome classes of the DC-S model was modelled using the Coxian phase-type distribution in addition to five other parametric distributions. Finally, the advantages of using the Waterfall SVM in the DC-S model as a resource for clinicians is discussed.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.728653  DOI: Not available
Share: