K-modestream algorithm for clustering categorical data streams

Clustering categorical data streams is a challenging problem because new data points are continuously adding to the already existing database at rapid pace and there exists no natural order among the categorical values. Recently, some algorithms have been discussed to tackle the problem of clustering the categorical data streams. However, in all these schemes the user needs to pre-specify the number of clusters, which is not trivial, and it renders to inefficient in the data stream environment. In this paper, we propose a new clustering algorithm, named it as k-modestream, which follows the k-modes algorithm paradigm to dynamically cluster the categorical data streams. It automatically computes the number of clusters and their initial modes simultaneously at regular time intervals. We analyse the time complexity of our scheme and perform various experiments using the synthetic and real world datasets to evaluate its efficacy.