A Divisive Clustering Algorithm for Performance Monitoring
of Large Networks using Maximum Common Subgraphs
R. VIJAYALAKSHMI
1
, R. NADARAJAN
1
, P. NIRMALA
1
, M. THILAGA
1
1
Department of Mathematics and Computer Applications
PSG College of Technology
Coimbatore, 641004, Tamil Nadu, India
rv@mca.psgtech.ac.in
ABSTRACT
In managing huge-enterprise communication networks, the ability to measure similarity is
an important performance monitoring function. It is possible to draw certain significant con-
clusions regarding effective utilization of networks by characterizing a computer network as
a time series of graphs with IP addresses as nodes and communication between nodes
as edges. Measuring similarity of graphs is a significant task in mining the graph data for
matching, comparing, and evaluating patterns in huge graph databases. The problem of
finding the nodes in the communication network which are always active can be formu-
lated as a Maximum Common Subgraph (MCS) detection problem. This paper presents a
Divisive Clustering MCS detection algorithm (DC-MCS) to find all maximum comomn sub-
graphs of k graphs in a graph database. The uniqueness of this algorithm lies in the facts
that it considers any number of input graphs can and it scans the graph database only
once. The series of experiments performed and the comparison of empirical results with
the existing algorithms further ensure the efficiency of the proposed algorithm.
Keywords: graph mining, graph similarity, graph matching, maximum common subgraph,
heap-based MCS algorithm.
Computing Classification System (CCS): H.2.8 Database Applications - Data mining,
I.5.3 Clustering - Algorithms, Similarity measures.
1 Introduction
Graph representation has been extensively used in modeling and investigating complicated
structural information such as circuits, images, chemical graphs, biological networks, the web,
XML documents and so on. Graph data mining refers to the extraction of novel and useful
knowledge referred as patterns from huge graph databases. Matching or comparing these
patterns is equivalent to determining the similarity among their graph representations (Vijay-
alakshmi et al., 2007). An important research activity that groups objects into different clusters
based on some measures of similarity is referred as cluster analysis. It is a data mining tech-
nique of unsupervised learning for statistical data analysis used in many fields, such as pattern
recognition, machine learning, image analysis and bioinformatics.
Graph matching refers to the problem of finding a mapping from nodes of a graph G
1
to the
International Journal of Artificial Intelligence,
ISSN 0974-0635; Int. J. Artif. Intell.
Autumn (October) 2011, Volume 7, Number A11
Copyright © 2011 by IJAI (CESER Publications)
www.ceserp.com/cp-jour
www.ceser.in/ijai.html
www.ceserpublications.com