An approach to determine clusters overlap for k-means clustering

Clustering is one of the major and interesting tools for many data analysis in business, science, medical, social network and other sources. Various clustering methods are available and applied in different fields, but unfortunately they are still limited especially when the data modelled are not completely in k disjoint partitions and may belong to multiple clusters. One of the ways to solve such a clustering problem or limitation is to find and calculate the overlapping proportions in order to better understand and model the data structure. Many overlapping approaches are proposed in the literature based on different methods, in this paper we using and extending the recent new approach based on incidental and proportion matrices related to forward and backward movement of the objects at different number of clusters when K-means method/algorithm is applied. The degree of separation and overlap between these clusters is evaluated, discussed and analysed through a simulated dataset.