Final published version, 14.7 MB, PDF document
Available under license: CC BY-NC: Creative Commons Attribution-NonCommercial 4.0 International License
Research output: Thesis › Doctoral Thesis
Research output: Thesis › Doctoral Thesis
}
TY - BOOK
T1 - Low-Density Cluster Separators for Large, High-Dimensional, Mixed and Non-Linearly Separable Data.
AU - Yates, Katie
PY - 2018
Y1 - 2018
N2 - The location of groups of similar observations (clusters) in data is a well-studied problem,and has many practical applications. There are a wide range of approaches to clustering,which rely on different definitions of similarity, and are appropriate for datasets with differentcharacteristics. Despite a rich literature, there exist a number of open problems inclustering, and limitations to existing algorithms.This thesis develops methodology for clustering high-dimensional, mixed datasets withcomplex clustering structures, using low-density cluster separators that bi-partition datasetsusing cluster boundaries that pass through regions of minimal density, separating regions ofhigh probability density, associated with clusters. The bi-partitions arising from a successionof minimum density cluster separators are combined using divisive hierarchical and partitionalalgorithms, to locate a complete clustering, while estimating the number of clusters.The proposed algorithms locate cluster separators using one-dimensional arbitrarily orientedsubspaces, circumventing the challenges associated with clustering in high-dimensionalspaces. This requires continuous observations; thus, to extend the applicability of the proposedalgorithms to mixed datasets, methods for producing an appropriate continuousrepresentation of datasets containing non-continuous features are investigated. The exactevaluation of the density intersected by a cluster boundary is restricted to linear separators.This limitation is lifted by a non-linear mapping of the original observations into a featurespace, in which a linear separator permits the correct identification of non-linearly separableclusters in the original dataset.In large, high-dimensional datasets, searching for one-dimensional subspaces, which resultin a minimum density separator is computationally expensive. Therefore, a computationallyefficient approach to low-density cluster separation using approximately optimalprojection directions is proposed, which searches over a collection of one-dimensional randomprojections for an appropriate subspace for cluster identification. The proposed approachesproduce high-quality partitions, that are competitive with well-established andstate-of-the-art algorithms.
AB - The location of groups of similar observations (clusters) in data is a well-studied problem,and has many practical applications. There are a wide range of approaches to clustering,which rely on different definitions of similarity, and are appropriate for datasets with differentcharacteristics. Despite a rich literature, there exist a number of open problems inclustering, and limitations to existing algorithms.This thesis develops methodology for clustering high-dimensional, mixed datasets withcomplex clustering structures, using low-density cluster separators that bi-partition datasetsusing cluster boundaries that pass through regions of minimal density, separating regions ofhigh probability density, associated with clusters. The bi-partitions arising from a successionof minimum density cluster separators are combined using divisive hierarchical and partitionalalgorithms, to locate a complete clustering, while estimating the number of clusters.The proposed algorithms locate cluster separators using one-dimensional arbitrarily orientedsubspaces, circumventing the challenges associated with clustering in high-dimensionalspaces. This requires continuous observations; thus, to extend the applicability of the proposedalgorithms to mixed datasets, methods for producing an appropriate continuousrepresentation of datasets containing non-continuous features are investigated. The exactevaluation of the density intersected by a cluster boundary is restricted to linear separators.This limitation is lifted by a non-linear mapping of the original observations into a featurespace, in which a linear separator permits the correct identification of non-linearly separableclusters in the original dataset.In large, high-dimensional datasets, searching for one-dimensional subspaces, which resultin a minimum density separator is computationally expensive. Therefore, a computationallyefficient approach to low-density cluster separation using approximately optimalprojection directions is proposed, which searches over a collection of one-dimensional randomprojections for an appropriate subspace for cluster identification. The proposed approachesproduce high-quality partitions, that are competitive with well-established andstate-of-the-art algorithms.
U2 - 10.17635/lancaster/thesis/204
DO - 10.17635/lancaster/thesis/204
M3 - Doctoral Thesis
PB - Lancaster University
CY - Lancaster
ER -