Clustering of Mixed Data by Integrating Fuzzy, Probabilistic, and Collaborative Clustering Framework

Clustering of numerical data is a very well researched problem and so is clustering of categorical data. However, when it comes to clustering of data with mixed attributes, the literature is not that rich. For numerical data, fuzzy clustering, in particular, the fuzzy c-means (FCM), is a very effective and popular algorithm, while for categorical data, use of mixture model is quite popular. In this paper, we propose a novel framework for clustering of mixed data which contains both numerical and categorical attributes. Our objective is to find the cluster substructures that are common to both the categorical and numerical data. Our formulation is inspired by the FCM algorithm (for dealing with numerical data), mixture models (for dealing with categorical data), and the collaborative clustering framework for aggregation of the two—it is an integrated approach that judiciously uses all three components. We use our algorithm on a few commonly used datasets and compare our results with those by some state of the art methods.