=Paper=
{{Paper
|id=Vol-1458/D04_CRC71_Hassani
|storemode=property
|title=None
|pdfUrl=https://ceur-ws.org/Vol-1458/D04_CRC71_Hassani.pdf
|volume=Vol-1458
}}
==None==
<pdf width="1500px">https://ceur-ws.org/Vol-1458/D04_CRC71_Hassani.pdf</pdf>
<pre>
   Multiple-Resolution Stream Clustering Using
               Graph Maintenance

              Marwan Hassani, Pascal Spaus, and Thomas Seidl

                 Data Management and Data Exploration Group
                      RWTH Aachen University, Germany
                  {hassani,spaus,seidl}@cs.rwth-aachen.de



      Abstract. Challenges for clustering streaming data are getting contin-
      uously more sophisticated. They are driven by stream properties such
      as the continuous data arrival, the time-critical processing of objects,
      the evolution of the data streams, the presence of outliers and the vary-
      ing densities of the data. Due to the continuously evolving nature of
      the stream, it is crucial that stream clustering algorithms autonomously
      detect clusters whose number, shapes and densities vary as the stream
      flows. We present the first hierarchical density-based stream clustering
      algorithm based on cluster stability, called HASTREAM [2] which is able
      to meet the above mentioned requirements.
      We show additionally, that HASTREAM inherited efficiency issues as
      the main drawback of density-based hierarchical clustering algorithms, as
      these were not the scope of its contribution. We present then
      I-HASTREAM [1], a first density-based hierarchical clustering algorithm
      that has considerably less computational time compared to the first pre-
      sented algorithm. I-HASTREAM utilizes and introduces techniques from
      the graph theory domain to devise an incremental update of the under-
      lying model instead of repeatedly performing the expensive calculations
      of the huge graph. Specifically the Prim’s algorithm for constructing
      the minimal spanning tree is adopted by introducing novel, incremental
      maintenance of the tree by vertex and edge insertion and deletion. The
      extensive experimental evaluation study on real world datasets shows
      that I-HASTREAM is considerably faster than HASTREAM while de-
      livering almost the same clustering quality.


References
1. Marwan Hassani, Pascal Spaus, Alfredo Cuzzocrea, and Thomas Seidl. Adap-
   tive stream clustering using incremental graph maintenance. In BigMine 2015 at
   KDD’15, pages 49–64, 2015.
2. Marwan Hassani, Pascal Spaus, and Thomas Seidl. Adaptive multiple-resolution
   stream clustering. In MLDM ’14, pages 134–148, 2014.

  Copyright c 2015 by the paper’s authors. Copying permitted only for private and
  academic purposes. In: R. Bergmann, S. Görg, G. Müller (Eds.): Proceedings of
  the LWA 2015 Workshops: KDML, FGWM, IR, and FGDB. Trier, Germany, 7.-9.
  October 2015, published at http://ceur-ws.org




                                        32

</pre>