Comparison of the Euclidean and linear discriminant functions in statistical discriminant analysis : a thesis presented to Massey University in partial fulfillment of the requirements for the degree of Master of Science in Operations Research at Massey University

Simple item page
dc.contributor.authorLim, Tiew-Kim
dc.date.accessioned2017-09-25T22:01:01Z
dc.date.available2017-09-25T22:01:01Z
dc.date.issued1992
dc.description.abstractIt is known that in the problem of statistical discriminant analysis, the linear discriminant function performs poorly when the dimension of the data, p, is large. It has been demonstrated by Marco, Young and Turner (1987) that the much simpler Euclidean distance classifier may out-perform the usual linear discriminant function under certain conditions. Their conclusions were arrived at from a simulation experiment which compared the probabilities of misclassification associated with the Euclidean distance classifier with those of the linear discriminant function, under certain conditions. In this dissertation, the asymptotic expansions of the probabilities of misclassification (the expected actual and expected plug-in error rates) associated with the two discriminant functions are obtained. These error rates are then used to investigate the relative performances of the two methods. Chapter 1 introduces the problem of discriminant analysis and describes the two competing procedures for discriminant analysis and some associated error rates. Then Chapter 2 reviews previous results, in the literature which show that the Euclidean distance classifier can perform better than the linear discriminant function. Chapter 3 gives the asymptotic expansions of the error rates, i.e. the expected actual error rate, and the expected plug-in error rate. The relative performances of the two methods on the basis of the asymptotic expansions are discussed in Chapter 4. The results show that in general the plug-in error rates for the Euclidean distance classifier give better estimates of the actual error rates for all dimensions of p which were considered, when compared to the linear discriminant function. Furthermore, the actual error rates for the Euclidean distance classifier also seem to give better estimates of the true error rates at large dimensions of p, when compared to the linear discriminant function. Certain situations where the linear discriminant function performs better than the Euclidean distance classifier are also identified. Final conclusions, discussions and recommendations for further work are given in Chapter 5.en_US
dc.identifier.urihttp://hdl.handle.net/10179/11990
dc.language.isoenen_US
dc.publisherMassey Universityen_US
dc.rightsThe Authoren_US
dc.subjectDiscriminant analysisen_US
dc.titleComparison of the Euclidean and linear discriminant functions in statistical discriminant analysis : a thesis presented to Massey University in partial fulfillment of the requirements for the degree of Master of Science in Operations Research at Massey Universityen_US
dc.typeThesisen_US
massey.contributor.authorLim, Tiew-Kim
thesis.degree.disciplineOperations Researchen_US
thesis.degree.grantorMassey Universityen_US
thesis.degree.levelMastersen_US
thesis.degree.nameMaster of Science (M. Sc.)en_US
Files
Original bundle
Now showing 1 - 2 of 2
Thumbnail Image
Name:
01_front.pdf
Size:
1.12 MB
Format:
Adobe Portable Document Format
Description:
Download
Thumbnail Image
Name:
02_whole.pdf
Size:
22.15 MB
Format:
Adobe Portable Document Format
Description:
Download
License bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
license.txt
Size:
3.32 KB
Format:
Item-specific license agreed upon to submission
Description:
Download
Collections
Theses and Dissertations