A Comparison Framework for Clustering Algorithms

Clustering means grouping target objects into different clusters in such a way, that each cluster contains similar objects and the objects in different clusters are dissimilar in a ceratin way. The main question of clustering is how to define the similarity and dissimilarity of the objects, and how to verify that the resulting clusters are good enough for a given purpose. There exist several methods for clustering that have different advantages and disadvantages regarding noise filtering, the shape of the resulting clusters or the number of clusters. Furthermore, validity indices are defined for measuring the goodness of the results. Comparing the clustering algorithms is a difficult task as there are many different approaches and they can have different cluster definitions as well. For this reason a novel comaprison approach is suggested in this paper that introduces a general clustering algorithm model. The main advantage of this model is that it gives a better overview of the clustering problem; furthermore, it divides the whole process into smaller and well separated substeps, that makes easier the investigation of the clustering algorithms.