Cluster Center Initialization for Categorical Data Using Multiple Attribute Clustering

The K-modes clustering algorithm is well known for its efficiency in clustering large categorical datasets. The K-modes algorithm requires random selection of initial cluster centers (modes) as seed, which leads to the problem that the clustering results are often dependent on the choice of initial cluster centers and non-repeatable cluster structures may be obtained. In this paper, we propose an algorithm to compute fixed initial cluster centers for the K-modes clustering algorithm that exploits a multiple clustering approach that determines cluster structures from the attribute values of given attributes in a data. The algorithm is based on the experimental observations that some of the data objects do not change cluster membership irrespective of the choice of initial cluster centers and individual attributes may provide some information about the cluster structures. Most of the time, attributes with few attribute values play significant role in deciding cluster membership of individual data object. The proposed algorithm gives fixed initial cluster center (ensuring repeatable clustering results), their computation is independent of the order of presentation of the data and has log-linear worst case time complexity with respect to the data objects. We tested the proposed algorithm on various categorical datasets and compared it against random initialization and two other available methods and show that it performs better than the existing methods.