On the connectivity of clusters

Abstract The existence of bridges or necks in a cluster indicates an undesirable property, the chaining effect, which often leads to misclassification in pattern recognition. To cope with this difficulty, a cluster analysis problem is formulated within the framework of graph theory. The study of the connectivity reveals the data structure of a given cluster. Bridges or necks are a consequence of poor connectedness of a cluster. A new algorithm containing the generalized Frisch labeling algorithm is developed in this paper to solve the chaining problem. The algorithm sequentially yields subclusters from a parent cluster when a splitting criterion is satisfied.