Adaptive stream clustering using incremental graph maintenance

M. Hassani, P. Spaus, A. Cuzzocrea, T. Seidl

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademicpeer-review

Abstract

Challenges for clustering streaming data are getting continuously more sophisticated. This trend is driven by the the emerging requirements of the application where those algorithms are used and the properties of the stream itself. Some of these properties are the continuous data arrival, the time-critical processing of objects, the evolution of the data streams, the presence of outliers and the varying densities of the data. Due to the fact that the stream evolves continuously in the process of its existence, it is crucial that the algorithm autonomously detects clusters of arbitrary shape, with different densities, and varying number of clusters. Recently, the first hierarchical density-based stream clustering algorithm based on cluster stability, called HASTREAM, was proposed. Although the algorithm was able to meet the above mentioned requirements, it inherited the main drawback of density-based hierarchical clustering algorithms, namely the efficiency issues. In this paper we propose I-HASTREAM, a first density-based hierarchical clustering algorithm that has considerably less computational time than HASTREAM. Our proposed method utilizes and introduces techniques from the graph theory domain to devise an incremental update of the underlying 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 a state-of-the-art hierarchical density-based stream clustering approach while delivering almost the same clustering quality.
Original languageEnglish
Title of host publicationWorkshop on Big Data, Streams and Heterogeneous Source Mining: Algorithms, Systems, Programming Models and Applications, 10 August 10, 2015
Pages49-64
Number of pages16
Publication statusPublished - 2015
Externally publishedYes
Event4th International Workshop on Big Data, Streams and Heterogeneous Source Mining: Algorithms, Systems, Programming Models and Applications (BigMine 2015) - Sydney, Australia
Duration: 10 Aug 201510 Aug 2015
Conference number: 4
http://dblp2.uni-trier.de/db/conf/kdd/bigmine2015

Publication series

NameJMLR Workshop and Conference Proceedings
PublisherJMLR
Volume41

Conference

Conference4th International Workshop on Big Data, Streams and Heterogeneous Source Mining: Algorithms, Systems, Programming Models and Applications (BigMine 2015)
Abbreviated titleBigMine2015
Country/TerritoryAustralia
CitySydney
Period10/08/1510/08/15
Internet address

Fingerprint

Dive into the research topics of 'Adaptive stream clustering using incremental graph maintenance'. Together they form a unique fingerprint.

Cite this