Hierarchical Clustering for Real-Time Stream Data with Noise

In stream data mining, stream clustering algorithms provide summaries of the relevant data objects that arrived in the stream. The model size of the clustering, i.e. the granularity, is usually determined by the speed (data per time) of the data stream. For varying streams, e.g. daytime or seasonal changes in the amount of data, most algorithms have to heavily restrict their model size such that they can handle the minimal time allowance. Recently the first anytime stream clustering algorithm has been proposed that flexibly uses all available time and dynamically adapts its model size. However, the method exhibits several drawbacks, as no noise detection is performed, since every point is treated equally, and new concepts can only emerge within existing ones. In this paper we propose the LiarTree algorithm, which is capable of anytime clustering and at the same time robust against noise and novelty to deal with arbitrary data streams.