Clustering with topological constraints

Dejan Grbec (2012) Clustering with topological constraints. EngD thesis.

Preview

PDF Download (1874Kb)

Abstract

In data analysis we often have to deal with object located somewhere in space. Data can be bound to cities, countries or other space object that have known spatial coordinates. This object can be clustered by selected attributes considering their adjacency. The object of diploma thesis was to develop procedures that consider objects neighboring relation. To determine the adjacency of object in space, we used the Voronoi diagram where the clusters of data should preset related parts of diagram. We have developed the application that provides hierarchical data clustering with different linkage types and different distance metrics. Application provides a graphical representation of Voronoi diagram, where the diagram elements are colored in cluster colors which enables us a better visual similarity presentation of items among themselves. During clustering we evaluate data division with validation method Silhouette, which give us the number of clusters where division is optimal.

Actions (login required)