Meta-Clustering Approach using Possibilistic Membership: Application to Retail Datasets

This paper proposes a new uncertain metaclustering approach devoted for handling uncertainty in the belonging of objects to different clusters with respect to their membership values. These values are presented under possibilistic framework to deal with uncertainty when an object belongs to several clusters. In addition, we use the meta-clustering which is based on the k-modes method to double-cluster categorical data. The meta-clustering is used to make an initial clustering of instances of a set and then, use the results to make a second clustering of another set. This double-clustering provides more meaningful clusters and more information to the user. Our metaclustering approach is developed and evaluated with the help of real-world retail datasets. The datasets consist of customers and products data relative to transactions made in a small retail store chain. We study and represent the uncertainty that can exist when a customer and a product belong to several clusters by the membership degrees.