Information Technology Reference
In-Depth Information
6
Conclusions
The article presents an approach to time complexity reduction in the process of cluster-
ing data. The idea is based on preparation of point-type input data to multidimensional
granules in the form of hyperboxes. Formation of the granules maximizes information
density transferred by the hyperboxes. The experiments showed the advantage of the
presented approach: significant time reduction of granulated data clustering in compar-
ison to point-type partitioning. It is particularly visible when data contain large number
of objects. Additionally, the quality of clustering result has not deteriorated when cop-
ing with granulated data, on the contrary - in most of the cases the quality has increased.
This is connected with the generalization ability of the presented method.
In case of SOSIG algorithm, clustering process can be performed on different res-
olution of data. Clustering of hyperboxes has been executed without changing the res-
olution. A three-level structure of data has been constructed by joining original point
(third down level) in hyperboxes (second level), whereas the top level contains divid-
ing of hyperboxes into clusters. Partitioning at the top level of hyperboxes granulation
(clustering) is composed of the same number of groups as partitioning point-type data.
The quality of created clusters is also comparable due to the similar values of quality
indices are similar.
The process of hyperbox creation is a type of aggregation operation, therefore the
major benefit of the presented method is shortening the time of cluster creation in com-
parison to the processing point-type data. It is particularly effective when data contain
large number of objects. Hyperboxes also determine additional level of relationship ex-
isting within data. Finally, the description of granules is more comprehensible since the
hyperboxes contain minimal and maximal values of attributes.
Acknowledgements.
The experiments have been performed on the computer cluster at
Faculty of Computer Science, Bialystok University of Technology.
This work was supported by Grant No. S/WI/5/08.
References
1. Bargiela, A., Pedrycz, W.: Classification and Clustering of Granular Data. In: IFSA World
Congress and 20th NAFIPS International Conference, vol. 3, pp. 1696-1701 (2001)
2. Bargiela, A., Pedrycz, W.: Granular Computing: an Introduction. Kluwer Academic Publish-
ers, Boston (2002)
3. Halkidi, M., Batistakis, Y.: On clustering validation techniques. Journal of Intelligent Infor-
mation Systems 17(2/3), 107-145 (2001)
4. Han, J., Kamber, M.: Data Mining: Concepts and Techniques. Morgan Kaufmann (2000)
5. Jain, A.K., Murty, M.N., Flynn, P.J.: Data clustering: a review. ACM Computing Sur-
veys 31(3), 264-323 (1999)
6. Kuzelewska, U.: Clustering with granular information processing. In: Proceedings of
ICAART Conference, vol. 1, pp. 89-97. SciTePress (2011)
7. Li, F., Xie, J., Xie, K.: Granular computing theory in the application of fault diagnosis. In:
Proceeding of Control and Decision Conference CCDC, pp. 595-597 (2008)
