Image and Video Analysis

Detecting and Verifying Dissimilar Patterns in Unlabelled Data

8th Online World Conference on Soft Computing in Industrial Applications, October 2003.

Clustering of unlabelled data is a difficult problem with numerous applications in various fields. When input space dimensions are many, the number of distinct patterns in the data is not known a priori, and feature scales are different, then the problem becomes much harder. In this paper we deal with such a problem. Our approach is based on an extension to hierarchical clustering that makes it suitable for data sets with numerous independent features. The results of this initial clustering are refined via a reclassification step. The issue of evaluation of hierarchical clustering methods is also discussed. The performance of the proposed methodology is demonstrated through the application to a synthetic data set and verified through application to a variety of well known machine learning data sets.

[ Bibtex ] [ PDF ]