A Novel MST based Multi-Prototype Clustering Algorithm

Clustering is the process of partitioning the data objects such that the degree of similarity among the objects in the same partition is more than the degree of dissimilarity between the objects of different classes. There are many applications of clustering like medicine, marketing, insurance, image analysis etc. The squared-error clustering algorithm uses single prototype to represent a cluster. However, the algorithm is incapable of detecting clusters with complex shapes. To resolve this issue, multi-prototype clustering algorithm has been proposed which can discover clusters of complex shapes, but it uses squared error clustering to produce prototypes. This paper proposes multi-prototype clustering algorithm based on Minimum Spanning Tree. Minimum Spanning Tree is known to be capable of detecting clusters with irregular boundaries. Two closest prototypes are merged based on Euclidean distance. At each step of merging, weighted standard deviation is calculated. The result is output for the set of clusters for which the weighted standard deviation is least. Experimental results for real data demonstrate the effectiveness of the proposed algorithm.