Clustering Algorithms for Data Stream

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
Volume: 04 Issue: 03 | Mar -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 1862
Clustering Algorithms for Data Stream
Karishma Nadhe1, Prof. P. M. Chawan2
1Student, Dept of CS & IT, VJTI Mumbai, Maharashtra, India
2Professor, Dept of CS & IT, VJTI Mumbai, Maharashtra, India
Abstract:
Nowadays streaming data is delivered by more and more applications, due to this crucial method for data and
knowledge engineering is considered to be clustering data streams. It is a two step process. A normal approach is to
summarize the data stream in real-time with an online process into so called micro-clusters. Local density estimates are
represented by micro-clusters by assembling the information of many data points which is defined in an area. A traditional
clustering algorithm is used in a second offline step, in which larger final clusters are formed by reclustering the micro-
clusters. For reclustering, the pseudo points which are used are actually coordinator of the micro-clusters with the weights
which are density estimates. However, in the online process, information about density in the area between micro-clusters is
not preserved and reclustering is based on possibly inaccurate assumptions about the distribution of data within and between
micro-clusters (e.g., uniform or Gaussian). In this paper we depicted various algorithms that are used for clustering and their
workings in brief.
Keywords: Data mining, data stream clustering, density-based clustering, micro-cluster, reclustering.
1. Introduction
Data mining is basically used to extract useful information from large sets of data. Clustering is the commonly
utilized data mining strategy. In this process, the classification of objects is done into different groups by partitioning sets
of data into a series of subsets or clusters. Detecting possible intrusions and also tracking the network data utilized to
identify changes in traffic patterns is the illustration of clustering. An additional requirement for this case is, as data is
produced it must be processed. The sequence of data points is ordered and unbounded in data stream[1],[2]. For
applications like GPS data from smart phones, web click-stream data, computer network monitoring data,
telecommunication connection data, readings from sensor nets, stock quotes, etc. such data streams are generated. The
reclustering approaches join micro-clusters which are separated by a small area of low density and also are close together.
These approaches completely ignore the data density in the area between the micro-clusters. To solve this problem, Tu
and Chen introduced an extension to the grid-based D-Stream algorithm based on the concept of attraction between
adjacent grids cells and showed its effectiveness.
And to solve this problem for micro- cluster-based algorithms we have discussed an algorithm. Here the concept
of a shared density graph has been introduced which explicitly captures the density of the original data between micro-

clusters during clustering and then it shows how the graph can be used for reclustering micro-clusters. The density in the
shared region between micro-clusters is estimated by this algorithm directly from the data.
2. Related Work
One of the well-researched areas is Density-based clustering and here we give a brief overview. We can see many
prototypical density-based clustering approaches like DBSCAN and several of its improvements[3]. In the first step the
density around each data point is estimated by DBSCAN by counting the number of points in a user speciﬁed eps-
neighbourhood and identifies core, border and noise points by applying user-speciﬁed thresholds. In a second step, core
points that are density-reachable are joined into a cluster. Finally, clusters are assigned by border points. Other
approaches are based on kernel density estimation e.g., DENCLUE or use shared nearest neighbours e.g., SNN,
CHAMELEON.
2.1 DBSCAN
To discover the clusters and the noise in a spatial database, the algorithm DBSCAN (Density Based Spatial
Clustering of Applications with Noise) is designed. To apply this algorithm the appropriate parameters Eps and MinPts of
each cluster and at least one point from the respective cluster should be known[4]. Then, using the correct parameters we
could retrieve all points that are density-reachable from the given point. To get this information in advance for all clusters
of the database there is no easy way. However, to determine the parameters Eps and MinPts of the "thinnest", i.e. least
dense, cluster in the database there is a simple and effective heuristic. For this reason, DBSCAN uses the same values for
Eps and MinPts for all clusters which are called the global values. For these global parameter values, the density
parameters of the “thinnest” cluster are considered to be good candidates specifying the lowest density which is not
considered to be noise.
DBSCAN starts with an arbitrary point p and retrieves all points that are density-reachable from p wrt. Eps and
MinPts. This is done to find a cluster. The result depends on p. This procedure yields a cluster wrt. Eps and MinPts if p is a
core point. If p is a border point, no points are density-reachable from p and DBSCAN visits the next point of the database.
Since we use global values for Eps and MinPts, if two clusters of different density are “close” to each other then DBSCAN
may merge two clusters into one cluster. Let the distance between two sets of points S1 and S2 be defined as dist (S1, S2) =
min {dist(p,q) | p∈ S1, q∈ S2}. Then, only if the distance between the two sets is larger than Eps two sets of points having at
least the density of the thinnest cluster will be separated from each other. Consequently, for the detected clusters with a
higher value for MinPts a recursive call of DBSCAN may be necessary.
In spatial databases for the task of class identification clustering algorithms are attractive. However, when applied
to large spatial databases these algorithms suffer from severe drawbacks.

2.2 DENCLUE
The DENCLUE algorithm employs a cluster model based on kernel density estimation. A cluster is defined by the
local maximum of the estimated density function. Data points going to the same local maximum are put into the same
cluster.
The DENCLUE algorithm works in two steps. In the first step map of the relevant portion of the data space is
constructed which is called a pre-clustering step. The map is used to speed up the calculation of the density function which
requires efficiently accessing of neighbouring portions of the data space. The actual clustering step is the second step, in
which the algorithm identifies the density-attractors and the corresponding density-attracted points.
New approach was introduced to clustering in large multimedia databases. They had formally introduced the
notion of influence and density functions as well as different notions of clusters which were both based on determining the
density-attractors of the density function. For an efficient implementation of their approach, they used a map-oriented
representation of the data they could efficiently determine a local density function which they introduced[5]. They had
shown the generality of their approach, i.e. that they were able to simulate a locality-based, partition-based, and
hierarchical clustering.
2.3 SNN
For low to medium dimensional data, density based algorithms such as DBSCAN, CLIQUE, MAFIA and DENCLUE
have shown to find clusters of different sizes and shapes, although not of different densities[6]. However, in high
dimensions, the notion of density is perhaps even more troublesome than that of distance. In particular, the traditional
Euclidean notion of density, the number of points per unit volume, becomes meaningless in high dimensions.
An alternative to a direct similarity is to define the similarity between a pair of points in terms of their shared
nearest neighbours. That is, the common or shared near neighbours confirms the similarity between two points. If point A
is close to point B and if they are both close to a set of points C then we can say that A and B are close with greater
confidence since their similarity is “confirmed” by the points in set C. This idea of shared nearest neighbour was first
introduced by Jarvis and Patrick. A similar idea was later presented in ROCK. In the Jarvis – Patrick scheme, a shared
nearest neighbour graph is constructed from the proximity matrix as follows[6]. Let i and j be two points. The strength of
the link between i and j is now defined as:
str(i, j) = ∑ (k +1− m) * (k +1− n) , where im = jn
In the equation above, k is the near neighbour list size, m and n are the positions of a shared near neighbour in i and j’s
lists. At this point, all edges with weights less than a user specified threshold are removed and all the connected
components in the resulting graph are our final clusters.

2.4 CHAMELEON
Chameleon is an agglomerative hierarchical clustering algorithm that overcomes the limitations of some
clustering algorithms. The key feature of Chameleon algorithm, to identifying the most similar pair of clusters it accounts
for both interconnectivity and closeness. Due to this it avoids the limitations that are experienced in other algorithms.
Furthermore, to model the degree of interconnectivity and closeness between each pair of clusters a novel approach is
used by Chameleon. The internal characteristics of the clusters are considered by the approach themselves. Thus, it can
automatically adapt to the internal characteristics of the merged clusters and it does not depend on a static, user-supplied
model. Chameleon operates on a sparse graph in which nodes represent data items, and weighted edges represent
similarities among the data items[7]. This sparse graph representation allows Chameleon to scale to large data sets and to
successfully use data sets that are available only in similarity space and not in metric spaces. Data sets in a metric space
have a fixed number of attributes for each data item, whereas data sets in a similarity space only provide similarities
between data items. Chameleon finds the clusters in the data set by using a two-phase algorithm. During the first phase,
Chameleon uses a graph-partitioning algorithm to cluster the data items into several relatively small sub-clusters. During
the second phase, it uses an algorithm to find the genuine clusters by repeatedly combining these sub-clusters.
2.5 DBSTREAM
This algorithm is supposed to be the first micro-cluster-based online clustering component that explicitly captures
the density between micro-clusters via a shared density graph. The density information in this graph is then exploited for
reclustering based on actual density between adjacent micro-clusters. We discuss the space and time complexity of
maintaining the shared density graph.
The basic idea of this work is that if we can capture not only the distance between two adjacent MCs but also the
connectivity using the density of the original data in the area between the MCs, then the reclustering results may be
improved. In the following we develop DBSTREAM which stands for density-based stream clustering.
2.5.1 Leader-Based Clustering
Leader-based clustering was introduced by Hartigan[8] as a conventional clustering algorithm. To apply the idea
to data streams it is considered straight-forward. DBSTREAM represents each MC by a leader which is a data point
defining the MC’s center and the density in an area of a user-specified radius r i.e. threshold around the center. This is
similar to DBSCAN’s concept of counting the points is an eps-neighbourhood, however, here the density is not estimated
for each point, but only for each MC which can easily be achieved for streaming data[3].
2.5.2 Competitive Learning
New leaders are chosen as points which cannot be assigned to an existing MC. The positions of these newly
formed MCs are most likely not ideal for the clustering. To remedy this problem, we use a competitive learning strategy
introduced in to move the MC centers towards each newly assigned point. To control the magnitude of the movement, we

use a neighbourhood function h() similar to self-organizing maps. In our implementation we use the popular Gaussian
neighbourhood function deﬁned between two points, a and b, as
With = r/3 indicating that the used neighbourhood size is +-3 standard deviations[1].
2.5.3 Capturing Shared Density
Capturing shared density directly in the online component is a new concept.
The shared density between two MCs, i and j, is estimated by , where sij is the shared weight and Aij is the size
of the overlapping area between the MCs[3].
A shared density graph Gsd(V,E) is an undirected weighted graph, where the set of vertices is the set of all MCs, i.e.,
V(Gsd)=MC, and the set of edges E(Gsd)={(vi,vj) | vi,vj V(Gsd) > 0} represents all the pairs of MCs for which
we have pairwise density estimates. Each edge is labeled with the pairwise density estimate .
In a typical clustering most MCs will not share density with each other. A very sparse shared density graph is the
result of this condition. We can exploit this fact for more efficient storage and manipulation of the graph. All fixed-radius
nearest-neighbours are needed to be found during clustering. Therefore, no additional increase in search time is incurred
while obtaining shared weights.
3. Conclusion
In this paper, we have analyzed some of the important algorithms which are required for clustering data streams.
We also discussed the first data stream clustering algorithm called DBSTREAM which explicitly records the density in the
area shared by micro-clusters and uses this data for reclustering. Experiments also show that shared-density reclustering
already performs extremely well when the online data stream clustering component is set to produce a small number of
large MCs. To achieve comparable results other popular reclustering strategies can only slightly improve over the results
of shared density reclustering and need significantly more MCs. Since it implies that we can tune the online component to
produce less micro-clusters for shared-density reclustering this is an important advantage. This improves performance
and, in many cases, the saved memory more than offset the memory requirement for the shared density graph.
REFERENCES
[1] C. Aggarwal, Data Streams:Models and Algorithms, (series Advances in Database Systems). NewYork, NY, USA:
Springer-Verlag, 2007.

[2] J. Gama, Knowledge Discovery from Data Streams, 1st ed. London, U.K.: Chapman & Hall, 2010.
[3] M. Hahsler and M. Bolanos, “Clustering Data Streams Based on Shared Density between Micro-Clusters,” IEEE
transactions on knowledge and data engineering, vol. 28, no. 6, june 2016.
[4] M. Ester, H.-P. Kriegel, J. Sander, and X. Xu, “A density-based algorithm for discovering clusters in large spatial
databases with noise,” in Proc. ACM SIGKDD Int. Conf. Knowl. Discovery Data Mining, 1996, pp. 226–231.
[5] A. Hinneburg, E. Hinneburg, and D. A. Keim, “An efﬁcient approach to clustering in large multimedia databases with
noise,” in Proc. 4th Int. Conf. Knowl. Discovery Data Mining, 1998, pp. 58–65.
[6] L. Ertoz, M. Steinbach, and V. Kumar, “A new shared nearest neighbour clustering algorithm and its applications,” in
Proc. Workshop Clustering High Dimensional Data Appl. 2nd SIAM Int. Conf. Data Mining, 2002, pp. 105–115.
[7] G. Karypis, E.-H. S. Han, and V. Kumar, “Chameleon: Hierarchical clustering using dynamic modeling,” Computer, vol. 32,
no. 8, pp. 68–75, Aug. 1999.
[8] J. A. Hartigan, “Clustering Algorithms”, 99th ed. New York, NY, USA: Wiley, 1975.

Clustering Algorithms for Data Stream

Recommended

More Related Content

What's hot (18)

Similar to Clustering Algorithms for Data Stream (20)

More from IRJET Journal (20)

Recently uploaded (20)

Clustering Algorithms for Data Stream