CETra: online cluster tracking for clustering of streaming data sources

Data stream clustering tasks may be applied to cluster streaming data objects (clustering by examples) or to cluster streaming data sources based on their temporal behavior (clustering by variables). We focus on the latter problem and propose CETra (Cluster evolution tracker)—the first online cluster tracking technique designed to provide information regarding cluster evolution in a streaming scenario of clustering by variables with efficient processing suitable for real-time problems. CETra can trace different intra and inter-cluster changes by considering not only statistics of interest but also the clusters’ membership, thus allowing a deeper understanding of the clustering results. Experimental evaluation using synthetic datasets and real data from meteorological sensors shows that CETra can track abrupt and gradual cluster transitions, while the competing method misses most of the gradual changes. Furthermore, CETra performs efficiently in a clustering environment for multiple streaming data sources, twice as fast as the related method.