Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.795977
Title: Uncertainty in discriminant analysis
Author: Hirst, David
Awarding Body: University of Glasgow
Current Institution: University of Glasgow
Date of Award: 1988
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
The aim of this thesis is to review and develop theory in discriminant analysis. In chapter one an example of medical diagnosis is considered, and two types of uncertainty are illustrated. Firstly, the log odds ratio can be close to zero, and secondly there can be considerable uncertainty about its true value. In chapter two we review existing methodology for constructing Interval estimates for the log odds when the two populations are normal. Five different methods are considered for distributions with equal covariances, and three are generalised to the unequal covariance situation. In chapter three these methods are Investigated by simulation. It is seen that only two methods in the equal covariance case give intervals of reliable empirical confidence, and only one generalises successfully to the unequal covariance case. In chapter four we go on to use the interval estimation methodology to assess a discriminant rule, suggesting some new ways of displaying the information available. In chapter five we develop the methods of chapter four to construct an accurate error rate estimator, which is compared with standard techniques by simulation. In chapter six the error rate estimator developed in chapter five is extended to the situation where there are more than two groups, and it is compared by simulation with generalisations of other standard techniques. The different methods are applied to a data set. In chapter seven the limitations of the work are discussed, and possible developments suggested.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.795977  DOI: Not available
Share: