Transportation Management Collage, Dalian Maritime University, Dalian, 116026, China

---

Abstract—Existing clustering ensemble algorithms for partitioning data need to know the generating process of clustering members clearly and most of them are not suitable to categorical data. In order to partition categorical data conveniently, at same time broaden the application of clustering ensemble, a fuzzy clustering ensemble algorithm was proposed in this paper, which not only can be used to classify categorical data, but also be used to combine results of multi clustering for numerical data or mixed categorical and numerical data. The proposed algorithm firstly made use of relationship degree between different attributes to prune part of attributes. Next, took the distribution of clustering members into account, Descartes subset and relationship degree between any two different objects were used for establishing the relationships between objects, which were under unsupervised circumstances and could get the minimum value of objective function of clustering and obtain corresponding optimal partitions. Then, choose the number of clusters satisfying the difference and differential rate of objective function local maximum as the optimal number of clusters and its corresponding partitions are optimal clustering. Finally, the proposed algorithm was applied in Synthesis dataset, Fellow-Small dataset, Zoo dataset, and results show the algorithm is effective and feasible.

Index Terms—fuzzy clustering; clustering ensemble; relationship degree; k-means clustering

[PDF]

Cite: Taoying Li and Yan Chen, " Fuzzy Clustering Ensemble with Selection of Number of Clusters," Journal of Computers vol. 5, no. 7, pp. 1112-1119, 2010.