Computational Intelligence - November 2017 - 98

Book
Book
Review
Review

A Primer on Cluster Analysis: 4
Basic Methods that (usually) Work,
by James C. Bezdek (First Edition
Design Publ., Sarasota, FL, 2017)

lmost everyone, a student, a medical
practitioner, an engineer, a researcher, or an astrophysicist encounters the concept of clustering for
some reason or other. Clustering was
important in the past, has become more
important in the present era of internet,
social networks, and big data, and will
continue to remain so in future. We need
clustering to find subgroups of cancers,
clusters of stars in our galaxies, to answer
web queries, to understand group
dynamics in social networks-the list
goes on. But what is a cluster? Oxford
Dictionary says a cluster is "a group of
similar things or people positioned or occurring
closely together". It looks very simple but
really not so when it comes to finding
clusters by computers. Why? First, except
in 2D or 3D, usually we shall not have
any idea about the number of clusters to
look for. Second, for a given problem
even if we somehow decide the number
of clusters, the other issues to resolve
include the choice of similarity, objective
function (model), algorithm to optimize
the objective function, and validation
method. For example, suppose we have a
set of points, which is more or less lying
on two circles of different centers and
radii. Then if a pair of points lies on
(close to) one of the two circles, the two
points are similar, though their Euclidean distance could be high. Here similarity is not defined by Euclidean
Digital Object Identifier 10.1109/MCI.2017.2742870
Date of publication: 12 October 2017

Nikhil R Pal
Indian Statistical Institute
Calcutta, INDIA

distance. On the other hand, for a dataset with points on two circular dishes,
the Euclidean distance is a good measure
of similarity/dissimilarity. More importantly, what human beings see as clusters (in
2D or 3D) or want to find as clusters, the
computer may not be able to see or find
the same clusters. The computer may find
a different set of clusters, which can be
equally valid but not agreeing with our
expectation. So there are two points of
view: computer view point and human
view point, and we expect both to agree.
The book, "A Primer on Cluster
Analysis: 4 Basic Methods that (usually)
Work" by James C Bezdek, is a distinguished one, which makes all these issues
very clear. This book
develops the necessary concepts such
as similarity, distance,
clusters, computer
view point, human
view point, and cluster validation in a
ver y log ical and
lucid manner with
plenty of easy-to-follow examples and
creative pictures.
Although the book
primarily focuses on
four types of popular
clustering algorithms,
it provides adequate materials and pointers for interested readers to sail through
a much wider family of clustering
algorithms. This book will be very useful (and of course enjoyable to read) to a
wide spectrum of readers including
beginners, researchers, and practitioners.
The book consists of 11 chapters divided (I love to call, clustered) into two
parts: Part I: The Art and Science of

IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE | NOVEMBER 2017

Clustering, which has five chapters and
Part II: Four Basic Models & Algorithms that contains the remaining six
chapters. The Part I introduces the necessary concepts that are shared by all clustering algorithms and dur ing this
course Prof. Bezdek clears many confusions that are often fac ed by students
and practitioners.
Chapter 1 deals with "the human
point of view" of clusters. It explains the
attributes of a set of objects that we use to
conceptualize or define clusters. Sometimes we use physical proximity of objects
to define clusters while in some other
cases we look for clusters in terms of
"similarity" of attributes that characterize
the objects. The author illustrates and
emphasizes how the
perception of similarity changes in a set
of objects depending
on how they are presented in relation to
their background.
This chapter clarifies how compactness and separation
lead to the concept
of clusters. If a computer needs to see
the same cluster as we
see, these ideas must
be carried over to the computer and this
issue is dealt with in Chapter 3.
Chapter 2 introduces some basics of
fuzzy sets so that readers who are not
familiar with fuzzy concepts can understand the materials on fuzzy clustering
and cluster validity presented in the latter
chapters. The author gives some compelling, yet easy-to-follow, examples to make
the difference between fuzzy modelling

Table of Contents for the Digital Edition of Computational Intelligence - November 2017