A Novel Three-Way Clustering Algorithm for Mixed-Type Data
Hong Zhi Yu, Zhihua Chang, Bing Zhou · 2017
Large quantities of mixed-type data, containing categorical, ordinal and numerical attributes, have commonly existed in real world. In this paper, a mixed-type data clustering method, which could deal with the uncertain situation that a cluster may not have a definite cluster boundary, is proposed inspired by the theory of three-way decisions. Many existing studies represent a cluster with a single set based on a two-way strategy, which does not adequately show the fact that a cluster may not have a well-defined cluster boundary. In this paper, we represent a cluster with a pair of sets, i.e., the core region and fringe region. The three-way clustering is suitable for dealing with uncertainty because it shows intuitively which objects are fringe to the cluster. Then, new measurements of distance between mixed-type data are proposed for different types of attribute values by means of a weighted tree structure. The measurement considers the semantic of attributes, the number of attribute values and the occurrence frequency of attribute values. Finally, a three-way clustering algorithm for mixed-type data is proposed. The experimental results show that the proposed distance measure of mixed-type data is reasonable and effective, the proposed algorithm is in a better performance at the accuracy and the average adjusted rand index than the compared algorithms in most cases.