Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

A stability analysis of sparse K-means

Apfel, Abraham (2017) A stability analysis of sparse K-means. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

[img]
Preview
PDF
Submitted Version

Download (1MB) | Preview

Abstract

Sparse K-Means clustering is an established method of simultaneously excluding uninformative features and clustering the observations. This is particularly useful in a high dimensional setting such as micro-array. However the subsets of features selected is often inaccurate when there are overlapping clusters, which adversely affects the clustering results. The current method also tends to be inconsistent, yielding high variability in the number of features selected.
We propose to combine a stability analysis with Sparse K-Means via performing Sparse K-Means on subsamples of the original data to yield accurate and consistent feature selection. After reducing the dimensions to an accurate, small subset of features, the standard K-Means clustering procedure is performed to yield accurate clustering results. Our method demonstrates improvement in accuracy and reduction in variability providing consistent feature selection as well as a reduction in the clustering error rate (CER) from the previously established Sparse K-Means clustering methodology. Our method continues to perform well in situations with strong cluster overlap where the previous methods were unsuccessful.
Public health significance: Clustering analysis on transcriptomic data has shown success in disease phenotyping and subgroup discovery. However, with current methodology, there is a lack of confidence in terms of the accuracy and reliability of the results, as they can be highly variable. With our methodology, we hope to allow the researcher to use cluster analysis to achieve disease phenotyping and subgroup discovery with confidence that they are uncovering accurate and stable results thus ensuring that their findings will allow reliable public health decisions to be made from their work.


Share

Citation/Export:
Social Networking:
Share |

Details

Item Type: University of Pittsburgh ETD
Status: Unpublished
Creators/Authors:
CreatorsEmailPitt UsernameORCID
Apfel, Abrahamaba44@pitt.eduaba440000-0003-4839-0979
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairAnderson, Stewartsja@pitt.edusja
Committee MemberTseng, Georgectseng@pitt.eductseng
Committee MemberLin, Yanyal14@pitt.eduyal14
Committee MemberTudorascu, Danadlt30@pitt.edudlt30
Date: 31 August 2017
Date Type: Publication
Defense Date: 5 May 2017
Approval Date: 31 August 2017
Submission Date: 7 May 2017
Access Restriction: 2 year -- Restrict access to University of Pittsburgh for a period of 2 years.
Number of Pages: 88
Institution: University of Pittsburgh
Schools and Programs: School of Public Health > Biostatistics
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: cluster analysis sparse k-means high-dimensional
Date Deposited: 31 Aug 2017 15:19
Last Modified: 01 Jul 2019 05:15
URI: http://d-scholarship.pitt.edu/id/eprint/32551

Metrics

Monthly Views for the past 3 years

Plum Analytics


Actions (login required)

View Item View Item