When a more robust clustering is either demanded by the user or by the application, the offline stage kicks off. Here the micro-clusters generated in the online steps are used as initial centers. A more rigorous clustering is performed in the background to generate the final list of clusters, called a macro-cluster. Since this is not time-critical, it can be executed the same way we work on data at rest.

For a more rigorous treatment of the various stream data clustering algorithm, please refer to State-of-the-art on clustering data streams Mohammed Ghesmoune, by Mustapha Lebbah and Hanene Azzag: https://bdataanalytics.biomedcentral.com/track/pdf/10.1186/s41044-016-0011-3?site=bdataanalytics.biomedcentral.com