218 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART A: SYSTEMS AND HUMANS, VOL. 38, NO.

1, JANUARY 2008
Automatic Clustering Using an Improved
Differential Evolution Algorithm
Swagatam Das, Ajith Abraham, Senior Member, IEEE, and Amit Konar, Member, IEEE
AbstractDifferential evolution (DE) has emerged as one of
the fast, robust, and efcient global search heuristics of current
interest. This paper describes an application of DE to the au-
tomatic clustering of large unlabeled data sets. In contrast to
most of the existing clustering techniques, the proposed algorithm
requires no prior knowledge of the data to be classied. Rather,
it determines the optimal number of partitions of the data on
the run. Superiority of the new method is demonstrated by
comparing it with two recently developed partitional clustering
techniques and one popular hierarchical clustering algorithm.
The partitional clustering algorithms are based on two powerful
well-known optimization algorithms, namely the genetic algorithm
and the particle swarm optimization. An interesting real-world
application of the proposed method to automatic segmentation of
images is also reported.
Index TermsDifferential evolution (DE), genetic algorithms
(GAs), particle swarm optimization (PSO), partitional clustering.
I. INTRODUCTION
C
LUSTERING means the act of partitioning an unlabeled
data set into groups of similar objects. Each group, called
a cluster, consists of objects that are similar between them-
selves and dissimilar to objects of other groups. In the past few
decades, cluster analysis has played a central role in a variety
of elds, ranging from engineering (e.g., machine learning,
articial intelligence, pattern recognition, mechanical engineer-
ing, and electrical engineering), computer sciences (e.g., web
mining, spatial database analysis, textual document collection,
and image segmentation), and life and medical sciences (e.g.,
genetics, biology, microbiology, paleontology, psychiatry, and
pathology) to earth sciences (e.g., geography, geology, and
remote sensing), social sciences (e.g., sociology, psychology,
archeology, and education), and economics (e.g., marketing and
business) [1][8].
Data clustering algorithms can be hierarchical or partitional
[9], [10]. Within each of the types, there exists a wealth of
subtypes and different algorithms for nding the clusters. In
hierarchical clustering, the output is a tree showing a sequence
Manuscript received April 13, 2006; revised September 23, 2006. The work
of A. Abraham was supported by the Centre for Quantiable Quality of Service
in Communication Systems (Q2S), Centre of Excellence, which is appointed
by the Research Council of Norway and funded by the Research Council,
Norwegian University of Science and Technology (NTNU), and UNINETT.
This paper was recommended by Associate Editor R. Subbu.
S. Das and A. Konar are with the Department of Electronics and Telecom-
munication Engineering, Jadavpur University, Kolkata 700032, India (e-mail:
swagatamdas19@yahoo.co.in; konaramit@yahoo.co.in).
A. Abraham is with the Q2S, Centre of Excellence, NTNU, 7491 Trondheim,
Norway (e-mail: ajith.abraham@ieee.org).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TSMCA.2007.909595
of clustering, with each cluster being a partition of the data set
[10]. Hierarchical algorithms can be agglomerative (bottom-up)
or divisive (top-down). Agglomerative algorithms begin with
each element as a separate cluster and merge them in succes-
sively larger clusters. Divisive algorithms begin with the whole
set and proceed to divide it into successively smaller clusters.
Hierarchical algorithms have two basic advantages [9]. First,
the number of classes need not be specied a priori, and second,
they are independent of the initial conditions. However, the
main drawback of hierarchical clustering techniques is that they
are static; that is, data points assigned to a cluster cannot move
to another cluster. In addition to that, they may fail to sepa-
rate overlapping clusters due to lack of information about the
global shape or size of the clusters [11]. Partitional clustering
algorithms, on the other hand, attempt to decompose the data
set directly into a set of disjoint clusters. They try to optimize
certain criteria (e.g., a square-error function, which is to be
detailed in Section II). The criterion function may emphasize
the local structure of the data, such as by assigning clusters to
peaks in the probability density function, or the global structure.
Typically, the global criteria involve minimizing some measure
of dissimilarity in the samples within each cluster while max-
imizing the dissimilarity of different clusters. The advantages
of the hierarchical algorithms are the disadvantages of the
partitional algorithms, and vice versa. An extensive survey of
various clustering techniques can be found in [11].
Clustering can also be performed in two different modes:
1) crisp and 2) fuzzy. In crisp clustering, the clusters are disjoint
and nonoverlapping in nature. Any pattern may belong to one
and only one class in this case. In fuzzy clustering, a pattern
may belong to all the classes with a certain fuzzy membership
grade [11]. The work described in this paper concerns crisp
clustering algorithms only.
The problem of partitional clustering has been approached
fromdiverse elds of knowledge, such as statistics (multivariate
analysis) [12], graph theory [13], expectationmaximization
algorithms [14], articial neural networks [15][17], evolu-
tionary computing [18], [19], and so on. Researchers all over
the globe are coming up with new algorithms, on a regular
basis, to meet the increasing complexity of vast real-world
data sets. Thus, it seems well nigh impossible to review the
huge and multifaceted literature on clustering in the scope of
this paper. We here, instead, conne ourselves to the eld of
evolutionary partitional clustering, where this paper attempts
to make a humble contribution. In the evolutionary approach,
clustering of a data set is viewed as an optimization problem
and solved by using an evolutionary search heuristic such as a
genetic algorithm (GA) [20], which is inspired by Darwinian
1083-4427/$25.00 2007 IEEE
DAS et al.: AUTOMATIC CLUSTERING USING AN IMPROVED DE ALGORITHM 219
evolution and genetics. The key idea is to create a population
of candidate solutions to an optimization problem, which is
iteratively rened by alteration and selection of good solu-
tions for the next iteration. Candidate solutions are selected
according to a tness function, which evaluates their quality
with respect to the optimization problem. In the case of GAs,
the alteration consists of mutation to explore solutions in the
local neighborhood of existing solutions and crossover to re-
combine information between different candidate solutions. An
important advantage of these algorithms is their ability to cope
with local optima by maintaining, recombining, and comparing
several candidate solutions simultaneously. In contrast, local
search heuristics, such as the simulated annealing algorithm
[21], only rene a single candidate solution and are notoriously
weak in coping with local optima. Deterministic local search,
which is used in algorithms like the K-means (to be introduced
in the next section) [12], [22], always converges to the nearest
local optimum from the starting position of the search.
Tremendous research effort has gone in the past few years to
evolve the clusters in complex data sets through evolutionary
computing techniques. However, not much research work has
been reported to determine the optimal number of clusters at the
same time. Most of the existing clustering techniques, based on
evolutionary algorithms, accept the number of classes K as an
input instead of determining the same on the run. Nevertheless,
in many practical situations, the appropriate number of groups
in a previously unhandled data set may be unknown or im-
possible to determine even approximately. For example, while
clustering a set of documents arising from the query to a search
engine, the number of classes K changes for each set of doc-
uments that result from an interaction with the search engine.
Also, if the data set is described by high-dimensional feature
vectors (which is very often the case), it may be practically im-
possible to visualize the data for tracking its number of clusters.
The objective of this paper is twofold. First, it aims at the
automatic determination of the optimal number of clusters in
any unlabeled data set. Second, it attempts to show that differ-
ential evolution (DE) [23], with a modication of the chromo-
some representation scheme, can give very promising results if
applied to the automatic clustering problem. DE is easy to im-
plement and requires a negligible amount of parameter tuning
to achieve considerably good search results. We modied the
conventional DE algorithm from its classical form to improve
its convergence properties. In addition to that, we used a novel
representation scheme for the search variables to determine the
optimal number of clusters. In this paper, we refer to the new
algorithm as the automatic clustering DE (ACDE) algorithm.
At this point, we would like to mention that the traditional ap-
proach of determining the optimal number of clusters in a data
set is using some specially devised statisticalmathematical
function (also known as a clustering validity index) to judge the
quality of partitioning for a range of cluster numbers. A good
clustering validity index is generally expected to provide global
minima/maxima at the exact number of classes in the data set.
Nonetheless, determination of the optimum cluster number us-
ing global validity measures is very expensive since clustering
has to be carried out for a variety of possible cluster numbers.
In the proposed evolutionary learning framework, a number of
trial solutions come up with different cluster numbers as well as
cluster center coordinates for the same data set. Correctness of
each possible grouping is quantitatively evaluated with a global
validity index (e.g., the CS or DavisBouldin (DB) measure
[35]). Then, through a mechanism of mutation and natural
selection, eventually, the best solutions start dominating the
population, whereas the bad ones are eliminated. Ultimately,
the evolution of solutions comes to a halt (i.e., converges)
when the ttest solution represents a near-optimal partitioning
of the data set with respect to the employed validity index.
In this way, the optimal number of classes along with the
accurate cluster center coordinates can be located in one runt
of the evolutionary optimization algorithm. A downside to the
proposed method is that its performance depends heavily on
the choice of a suitable clustering validity index. An inefcient
validity index may result into many false clusters (due to the
overtting of data) even when the actual number of clusters
in the given data set may be very much tractable. However,
with a judicious choice of the validity index, the proposed
algorithm can automate the entire process of clustering and
yield near-optimal partitioning of any previously unhandled
data set in a reasonable amount of time. This is certainly a very
desirable feature of a real-life pattern recognition task.
We have extensively compared the ACDE with two other
state-of-the-art automatic clustering techniques [24], [25] based
on GA and particle swarm optimization (PSO) [26]. In addition,
the quality of the nal solutions has been compared with a
standard agglomerative hierarchical clustering technique. The
following performance metrics have been used in the com-
parative analysis: 1) the accuracy of nal clustering results;
2) the speed of convergence; and 3) the robustness (i.e., ability
to produce nearly same results over repeated runs). The test suit
chosen for this paper consists of ve real-life data sets. Finally,
an interesting application of the proposed algorithm has been
illustrated with reference to the automatic segmentation of a
few well-known grayscale images.
The rest of this paper is organized as follows. Section II
denes the clustering problem in a formal language and gives
a brief overview of a previous work done in the eld of evolu-
tionary partitional clustering. Section III outlines the proposed
ACDE algorithm. Section IV describes the ve real data sets
used for experiments, the simulation strategy, the algorithms
used for comparison, and their parameter setup. Results of clus-
tering over ve real-life data sets and an application in image
pixel classication are presented in Section V. Conclusions are
provided in Section VI.
II. SCIENTIFIC BACKGROUNDS
A. Problem Denition
A pattern is a physical or abstract structure of objects. It
is distinguished from others by a collective set of attributes
called features, which together represent a pattern [27]. Let
P = P
1
, P
2
, . . . , P
n
be a set of n patterns or data points,
each having d features. These patterns can also be represented
by a prole data matrix X
nd
with n d-dimensional row
vectors. The ith row vector

X
i
characterizes the ith object from
the set P, and each element X
i,j
in

X
i
corresponds to the
220 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART A: SYSTEMS AND HUMANS, VOL. 38, NO. 1, JANUARY 2008
jth real-value feature (j = 1, 2, . . . , d) of the ith pattern (i =
1, 2, . . . , n). Given such an X
nd
matrix, a partitional cluster-
ing algorithm tries to nd a partition C = C
1
, C
2
, . . . , C
K

of K classes, such that the similarity of the patterns in the same

cluster is maximumand patterns fromdifferent clusters differ as
far as possible. The partitions should maintain three properties.
1) Each cluster should have at least one pattern assigned,
i.e., C
i
,= i 1, 2, . . . , K.
2) Two different clusters should have no pattern in common,
i.e., C
i
C
j
= i ,= j and i, j 1, 2, . . . , K.
3) Each pattern should denitely be attached to a cluster i.e.,

K
i=1
C
i
= P.
Since the given data set can be partitioned in a number of
ways, maintaining all of the aforementioned properties, a tness
function (some measure of the adequacy of the partitioning)
must be dened. The problem then turns out to be one of
nding a partition C

of optimal or near-optimal adequacy,

as compared to all other feasible solutions C = C
1
, C
2
, . . . ,
C
N(n,K)
, where
N(n, K) =
1
K!
K

i=1
(1)
i
_
K
i
_
i
(K i)
i
(1)
is the number of feasible partitions. This is the same as
Optimize f(X
nd
, C)
C (2)
where C is a single partition from the set C, and f is a
statisticalmathematical function that quanties the goodness
of a partition on the basis of the distance measure of the patterns
(please see Section II-C). It has been shown in [28] that the
clustering problem is NP-hard when the number of clusters
exceeds 3.
B. Similarity Measures
As previously mentioned, clustering is the process of recog-
nizing natural groupings or clusters in multidimensional data
based on some similarity measures. Hence, dening an appro-
priate similarity measure plays a fundamental role in clustering
[11]. The most popular way to evaluate similarity between two
patterns amounts to the use of a distance measure. The most
widely used distance measure is the Euclidean distance, which
between any two d-dimensional patterns

X
i
and

X
j
is given by
d(

X
i
,

X
j
) =

_
d

p=1
(X
i,p
X
j,p
)
2
= |

X
i


X
j
|. (3)
The Euclidean distance measure is a special case (when
= 2) of the Minowsky metric [11], which is dened as
d

X
i
,

X
j
)=
_
d

p=1
(X
i,p
X
j,p
)

_
1/
=|

X
i


X
j
|

. (4)
When = 1, the measure is known as the Manhattan dis-
tance [28].
The Minowsky metric is usually not efcient for clustering
data of high dimensionality, as the distance between the pat-
terns increases with the growth of dimensionality. Hence, the
concepts of near and far become weaker [29]. Furthermore,
according to Jain et al. [11], for the Minowsky metric, the large-
scale features tend to dominate over the other features. This can
be solved by normalizing the features over a common range.
One way to do the same is by using the cosine distance (or
vector dot product), which is dened as

X
i
,

X
j
) =
d

p=1
X
i,p
X
j,p
|

X
i
||

X
j
|
. (5)
The cosine distance measures the angular difference of the
two data vectors (patterns) and not the difference of their
magnitudes. Another distance measure that needs mention in
this context is the Mahalanabis distance, which is dened as
d
M
(

X
i
,

X
j
) = (

X
i


X
j
)
1
(

X
i


X
j
) (6)
where is the covariance matrix of the patterns. The
Mahalanabis distance assigns different weights to different
features based on their variances and pairwise linear corre-
lations [11].
C. Clustering Validity Indexes
Cluster validity indexes correspond to the statistical
mathematical functions used to evaluate the results of a clus-
tering algorithm on a quantitative basis. Generally, a cluster
validity index serves two purposes. First, it can be used to
determine the number of clusters, and second, it nds out
the corresponding best partition. One traditional approach for
determining the optimum number of classes is to repeatedly
run the algorithm with a different number of classes as input
and then to select the partitioning of the data resulting in the
best validity measure [30]. Ideally, a validity index should take
care of the two aspects of partitioning.
1) Cohesion: The patterns in one cluster should be as similar
to each other as possible. The tness variance of the
patterns in a cluster is an indication of the clusters
cohesion or compactness.
2) Separation: Clusters should be well separated. The dis-
tance among the cluster centers (may be their Euclidean
distance) gives an indication of cluster separation.
For crisp clustering, some of the well-known indexes avail-
able in the literature are the Dunns index (DI) [31], the
CalinskiHarabasz index [32], the DB index [33], the Pakhira
Bandyopadhyay Maulik (PBM) index [34], and the CS measure
[35]. All these indexes are optimizing in nature, i.e., the maxi-
mum or minimum values of these indexes indicate the appropri-
ate partitions. Because of their optimizing character, the cluster
validity indexes are best used in association with any optimiza-
tion algorithm such as GA, PSO, etc. In what follows, we will
discuss only two validity measures in detail, which have been
employed in the study of our automatic clustering algorithm.
1) DB Index: This measure is a function of the ratio of the
sum of within-cluster scatter to between-cluster separation, and
DAS et al.: AUTOMATIC CLUSTERING USING AN IMPROVED DE ALGORITHM 221
it uses both the clusters and their sample means. First, we dene
the within ith cluster scatter and the between ith and jth cluster
distance, respectively, i.e.,
S
i,q
=
_
_
1
N
i

XC
i
|

X m
i
|
q
2
_
_
1/q
(7)
d
ij,t
=
_
d

p=1
[m
i,p
m
j,p
[
t
_
1/t
= | m
i
m
j
|
t
(8)
where m
i
is the ith cluster center, q, t 1, q is an integer,
and q and t can be independently selected. N
i
is the number
of elements in the ith cluster C
i
. Next, R
i,qt
is dened as
R
i,qt
= max
jK,j,=i
_
S
i,q
+S
j,q
d
ij,t
_
. (9)
Finally, we dene the DB measure as
DB(K) =
1
K
K

i=1
R
i,qt
. (10)
The smallest DB(K) index indicates a valid optimal
partition.
2) CS Measure: Recently, Chou et al. have proposed the
CS measure [35] for evaluating the validity of a clustering
scheme. Before applying the CS measure, the centroid of a
cluster is computed by averaging the data vectors that belong
to that cluster using
m
i
=
1
N
i

x
j
C
i
x
j
. (11)
A distance metric between any two data points

X
i
and

X
j
is
denoted by d(

X
i
,

X
j
). Then, the CS measure can be dened as
CS(K) =
1
K
K

i=1
_
1
N
i

X
i
C
i
max

X
q
C
i
_
d(

X
i
,

X
q
)
_
_
1
K
K

i=1
_
min
jK,j,=i
d( m
i
, m
j
)
_
=
K

i=1
_
1
N
i

X
i
C
i
max

X
q
C
i
_
d(

X
i
,

X
q
)
_
_
K

i=1
_
min
jK,j,=i
d( m
i
, m
q
)
_ . (12)
As can easily be perceived, this measure is a function of
the ratio of the sum of within-cluster scatter to between-cluster
separation and has the same basic rationale as the DI and DB
measures. According to Chou et al., the CS measure is more
efcient in tackling clusters of different densities and/or sizes
than the other popular validity measures, the price being paid in
terms of high computational load with increasing K and n.
D. Brief Review of the Existing Works
The most widely used iterative K-means algorithm [22] for
partitional clustering aims at minimizing the intracluster spread
(ICS), which for K cluster centers can be dened as
ICS(C
1
, C
2
, . . . , C
K
) =
K

i=1

X
i
C
i
|

X
i
m
i
|
2
. (13)
The K-means (or hard C-means) algorithm starts with K
cluster centroids (these centroids are initially randomly selected
or derived from some a priori information). Each pattern in
the data set is then assigned to the closest cluster center. The
centroids are updated by using the mean of the associated
patterns. The process is repeated until some stopping criterion
is met. The K-means has two main advantages [11].
1) It is very easy to implement.
2) The time complexity is only O(n) (n being the number of
data points), which makes it suitable for large data sets.
However, it suffers from three disadvantages.
1) The user has to specify in advance the number of classes.
2) The performance of the algorithm is data dependent.
3) The algorithm uses a greedy approach and is heavily
dependent on the initial conditions. This often leads
K-means to converge to suboptimal solutions.
The remaining paragraphs of this section provide a summary
of the most important applications of evolutionary computing
techniques to the partitional clustering problem.
The rst application of GAs to clustering was introduced by
Raghavan and Birchand [36], and it was the rst approach of
using a direct encoding of the objectcluster association. The
idea in this approach is to use a genetic encoding that directly
allocates n objects to K clusters, such that each candidate
solution consists of n genes, each with an integer value in the
interval [1, K]. For example, for n = 5 and K = 3, the encod-
ing 11322 allocates the rst and second objects to cluster 1,
the third object to cluster 3, and the fourth and fth objects to
cluster 2; thus, the clusters (1, 2, 3, 4, 5) are identied.
Based on this problem representation, the GA tries to nd the
optimal partition according to a tness function that measures
the partition goodness. It has been shown that such an algorithm
outperforms K-means in the analysis of simulated and real
data sets (e.g., [37]). However, the representation scheme has
a major drawback because of its redundancy; for instance,
11322 and 22311 represent the same grouping solution
(1, 2, 3, 4, 4). Falkenauer [18] tackled this problem in an
elegant way: in addition to the encoding of n genes representing
each objectcluster association, they represent the group labels
as additional genes in the encoding and apply ad hoc evolution-
ary operators on them.
The second kind of GA approach to partitional clustering is
to encode cluster-separating boundaries. Bandyopadhyay et al.
[38] used GAs to determine hyperplanes as decision bound-
aries, which divide the attribute feature space to separate the
clusters. For this, they encode the location and orientation of
a set of hyperplanes with a gene representation of exible
length. Apart from minimizing the number of misclassied
222 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART A: SYSTEMS AND HUMANS, VOL. 38, NO. 1, JANUARY 2008
objects, their approach tries to minimize the number of required
hyperplanes. The third way to use GAs in partitional clustering
is to encode a representative variable (typically a centroid or
medoid) and, optionally, a set of parameters to describe the
extent and shape of the variance for each cluster. Srikanth et al.
[39] proposed an approach that encodes the center, extend, and
orientation of an ellipsoid for each cluster.
Some researchers introduced hybrid clustering algorithms,
combining classical clustering techniques with GAs [40]. For
example, Krishna and Murty [41] introduced a GA with di-
rect encoding of objectcluster associations as in [39], but
applied K-means to determine the quality of the GA candidate
solutions. Kuo et al. [42] used adaptive resonance theory 2
(ART2) neural network to determine an initial solution and then
applied genetic K-means algorithm to nd the nal solution
for analyzing Web-browsing paths in electronic commerce. The
proposed method was also compared with ART2, followed by
K-means.
Finding an optimal number of clusters in a large data set is
usually a challenging task. The problem has been investigated
by several researchers [43], [44], but the outcome is still un-
satisfactory [45]. Lee and Antonsson [46] used an evolutionary
strategy (ES) [47]-based method to dynamically cluster a data
set. The proposed ES implemented variable-length individuals
to search for both centroids and optimal number of clusters.
An approach to dynamically classify a data set using evolu-
tionary programming [48] can be found in [49], where two
tness functions are simultaneously optimized: one gives the
optimal number of clusters, whereas the other leads to a proper
identication of each clusters centroid. Bandyopadhyay et al.
[24] devised a variable string-length genetic algorithm to
tackle the dynamic clustering problem using a single tness
function.
Recently, researchers working in this area have started taking
some interest on two promising approaches to numerical opti-
mization, namely the PSOand the DE. Paterlinia and Krink [50]
used a DE algorithm and compared its performance with a PSO
and a GA algorithm over the partitional clustering problem.
Their work is focused on nonautomatic clustering with a pre-
assigned number of clusters. In [51], Omran et al. proposed an
image segmentation algorithmbased on the PSO. The algorithm
nds the centroids of a user-specied number of clusters, where
each cluster groups together the similar pixels. They used a
crisp criterion function for evaluating the partitions on the
image data. Very recently, the same authors have come up with
another automatic hard clustering scheme [25]. The algorithm
starts by partitioning the data set into a relatively large number
of clusters to reduce the effect of the initialization. Using a
binary PSO [52], an optimal number of clusters is selected.
Finally, the centroids of the chosen clusters are rened through
the K-means algorithm. The authors applied the algorithm for
segmentation of natural, synthetic, and multispectral images.
Omran et al. also devised a nonautomatic crisp clustering
scheme based on DE and illustrated the application of the
algorithm to image segmentation problems in [53]. However,
to the best of our knowledge, DE has not been applied to the
automatic clustering of large real-life data sets as well as image
pixels until date.
III. DE-BASED AUTOMATIC CLUSTERING
A. Classical DE Algorithm and Its Modication
The classical DE [23] is a population-based global
optimization algorithm that uses a oating-point (real-coded)
representation. The ith individual vector (chromosome) of
the population at time-step (generation) t has d components
(dimensions), i.e.,

Z
i
(t) = [Z
i,1
(t), Z
i,2
(t), . . . , Z
i,d
(t)] . (14)
For each individual vector

Z
k
(t) that belongs to the current
population, DE randomly samples three other individuals, i.e.,

Z
i
(t),

Z
j
(t), and

Z
m
(t), from the same generation (for dis-
tinct k, i, j, and m). It then calculates the (componentwise)
difference of

Z
i
(t) and

Z
j
(t), scales it by a scalar F (usually
[0, 1]), and creates a trial offspring

U
i
(t + 1) by adding the
result to

Z
m
(t). Thus, for the nth component of each vector
U
k,n
(t + 1)
=
_
Z
m,n
(t) +F(Z
i,n
(t)Z
j,n
(t)) , if rand
n
(0, 1)<Cr
Z
k,n
(t), otherwise.
(15)
Cr [0, 1] is a scalar parameter of the algorithm, called the
crossover rate. If the new offspring yields a better value of the
objective function, it replaces its parent in the next generation;
otherwise, the parent is retained in the population, i.e.,

Z
i
(t + 1)=
_
_
_

U
i
(t + 1), if f
_

U
i
(t + 1)
_
>f
_

Z
i
(t)
_

Z
i
(t), if
_

U
i
(t + 1)
_
f
_

Z
i
(t)
_
(16)
where f() is the objective function to be maximized.
To improve the convergence properties of DE, we have tuned
its parameters in two different ways here. In the original DE, the
difference vector (

Z
i
(t)

Z
j
(t)) is scaled by a constant factor
F. The usual choice for this control parameter is a number
between 0.4 and 1. We propose to vary this scale factor in a
random manner in the range (0.5, 1) by using the relation
F = 0.5 (1 +rand(0, 1)) (17)
where rand(0, 1) is a uniformly distributed random number
within the range [0, 1]. The mean value of the scale factor
is 0.75. This allows for stochastic variations in the amplica-
tion of the difference vector and thus helps retain population
diversity as the search progresses. In [54], we have already
shown that the DE with random scale factor (DERANDSF) can
meet or beat the classical DE and also some versions of the
PSO in a statistically signicant manner. In addition to that,
here, we also linearly decrease the crossover rate Cr with time
from Cr
max
= 1.0 to Cr
min
= 0.5. If Cr = 1.0, it means that all
components of the parent vector are replaced by the difference
vector operator according to (12). However, at the later stages
of the optimizing process, if Cr is decreased, more components
of the parent vector are then inherited by the offspring. Such
a tuning of Cr helps exhaustively explore the search space at
DAS et al.: AUTOMATIC CLUSTERING USING AN IMPROVED DE ALGORITHM 223
Fig. 1. Chromosome encoding scheme in the proposed method. A total of ve cluster centers have been encoded for a 3-D data set. Only the activated cluster
centers have been shown as orange circles.
the beginning but nely adjust the movements of trial solutions
during the later stages of search, so that they can explore the
interior of a relatively small space in which the suspected global
optimum lies. The time variation of Cr may be expressed in the
form of the following equation:
Cr = (Cr
max
Cr
min
) (MAXIT iter)/MAXIT (18)
where Cr
max
and Cr
min
are the maximum and minimum values
of crossover rate Cr, respectively; iter is the current iteration
number; and MAXIT is the maximum number of allowable
iterations.
B. Chromosome Representation
In the proposed method, for n data points, each d dimen-
sional, and for a user-specied maximum number of clusters
K
max
, a chromosome is a vector of real numbers of dimension
K
max
+K
max
d. The rst K
max
entries are positive oating-
point numbers in [0, 1], each of which controls whether the
corresponding cluster is to be activated (i.e., to be really used
for classifying the data) or not. The remaining entries are
reserved for K
max
cluster centers, each d dimensional. For
example, the velocity vector

V
i
(t) of the ith chromosome is
shown in the equation at the bottom of the page.
The jth cluster center in the ith chromosome is active or
selected for partitioning the associated data set if T
i,j
> 0.5. On
the other hand, if T
i,j
< 0.5, the particular jth cluster is inactive
in the ith chromosome. Thus, the T
i,j
s behave like control
genes (we call them activation thresholds) in the chromosome
governing the selection of the active cluster centers. The rule
for selecting the actual number of clusters specied by one
chromosome is
IF T
i,j
> 0.5, THENthe jth cluster center
m
i,j
is ACTIVE
ELSE m
i,j
is INACTIVE. (19)
As an example, consider the chromosome encoding scheme
in Fig. 1. There are at most ve 3-D cluster centers, among
which, according to the rule presented in (19), the second
(6, 4.4, 7), third (5.3, 4.2, 5), and fth ones (8, 4, 4) have
been activated for partitioning the data set. The quality of the
partition yielded by such a chromosome can be judged by an
appropriate cluster validity index.
When a new offspring chromosome is created according to
(15) and (16), at rst, the T values are used to select [using (19)]
the active cluster centroids. If due to mutation some threshold
T
i,j
in an offspring exceeds 1 or becomes negative, it is force-
fully xed to 1 or 0, respectively. However, if it is found that no
ag could be set to 1 in a chromosome (all activation thresholds
are smaller than 0.5), we randomly select two thresholds and
reinitialize them to a random value between 0.5 and 1.0. Thus,
the minimum number of possible clusters is 2.

V
i
(t) = T
i,1
T
i,2
. . . T
i
, K
max
. .
Activation Threshhold
m
i,1
m
i,2
. . . m
i,K
max
. .
Cluster Centroids
224 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART A: SYSTEMS AND HUMANS, VOL. 38, NO. 1, JANUARY 2008
C. Fitness Function
One advantage of the ACDE algorithm is that it can use
any suitable validity index as its tness function. Here, we
conducted two different sets of experiments with two different
tness functions. These two functions are built on two clus-
tering validity measures, namely the CS measure and the DB
measure (refer to Sections II-C1 and C2). The CS-measure-
based tness functions can be described as
f
1
=
1
CS
i
(K) + eps
. (20)
Similarly, we may express the DB-index-based tness
function as
f
2
=
1
DB
i
(K) + eps
(21)
where DB
i
is the DB index, which is evaluated on the partitions
yielded by the ith chromosome (or the ith particle for PSO), and
eps is the same as before.
D. Avoiding Erroneous Chromosomes
There is a possibility that, in our scheme, during computation
of the CS and/or DB measures, a division by zero may be
encountered. This may occur when one of the selected cluster
centers is outside the boundary of distributions of the data set.
To avoid this problem, we rst check to see if any cluster has
fewer than two data points in it. If so, the cluster center positions
of this special chromosome are reinitialized by an average
computation. We put n/K data points for every individual
cluster center, such that a data point goes with a center that is
nearest to it.
E. Pseudocode of the ACDE Algorithm
The pseudocode for the complete ACDE algorithm is
given here.
Step 1) Initialize each chromosome to contain K number of
randomly selected cluster centers and K (randomly
chosen) activation thresholds in [0, 1].
Step 2) Find out the active cluster centers in each chromo-
some with the help of the rule described in (19).
Step 3) For t = 1 to t
max
do
a) For each data vector

X
p
, calculate its distance
metric d(

X
p
, m
i,j
) from all active cluster centers
of the ith chromosome

V
i
.
b) Assign

X
p
to that particular cluster center m
i,j
,
where
d(

X
p
, m
i,j
) = min
b1,2,...,K
_
d(

X
p
, m
i,b
)
_
.
c) Check if the number of data points that belong to
any cluster center m
i,j
is less than 2. If so, update
the cluster centers of the chromosome using the
concept of average described earlier.
d) Change the population members according to
the DE algorithm outlined in (15)(18). Use the
tness of the chromosomes to guide the evolution
of the population.
Step 4) Report as the nal solution the cluster centers and
the partition obtained by the globally best chromo-
some (one yielding the highest value of the tness
function) at time t = t
max
.
IV. EXPERIMENTS AND RESULTS FOR
THE REAL-LIFE DATA SETS
In this section, we compare performance of the ACDE algo-
rithm with two recently developed partitional clustering algo-
rithms and one standard hierarchical agglomerative clustering
based on the linkage metric of average link [55]. The former
two algorithms are well known as the genetic clustering with an
unknown number of clusters K (GCUK) [24] and the dynamic
clustering PSO (DCPSO) [25]. Moreover, to investigate the
effects of the changes made in the classical DE algorithm,
we have compared the ACDE with an ordinary DE-based
clustering method, which uses the same chromosome represen-
tation scheme and tness function as the ACDE. The classical
DE scheme that we have used is referred in the literature as
the DE/rand/1/bin [23], where bin stands for the binomial
crossover method.
A. Data Sets Used
The following real-life data sets [56], [57] are used in this
paper. Here, n is the number of data points, d is the number of
features, and K is the number of clusters.
1) Iris plants database (n = 150, d = 4, K = 3): This is
a well-known database with 4 inputs, 3 classes, and
150 data vectors. The data set consists of three different
species of iris ower: Iris setosa, Iris virginica, and
Iris versicolour. For each species, 50 samples with four
features each (sepal length, sepal width, petal length, and
petal width) were collected. The number of objects that
belong to each cluster is 50.
2) Glass (n = 214, d = 9, K = 6): The data were sam-
pled from six different types of glass: 1) building win-
dows oat processed (70 objects); 2) building windows
nonoat processed (76 objects); 3) vehicle windows
oat processed (17 objects); 4) containers (13 objects);
5) tableware (9 objects); and 6) headlamps (29 ob-
jects). Each type has nine features: 1) refractive index;
2) sodium; 3) magnesium; 4) aluminum; 5) silicon;
6) potassium; 7) calcium; 8) barium; and 9) iron.
3) Wisconsin breast cancer data set (n = 683, d=9, K=2):
The Wisconsin breast cancer database contains nine rele-
vant features: 1) clump thickness; 2) cell size uniformity;
3) cell shape uniformity; 4) marginal adhesion; 5) single
epithelial cell size; 6) bare nuclei; 7) bland chromatin;
8) normal nucleoli; and 9) mitoses. The data set has two
classes. The objective is to classify each data vector into
benign (239 objects) or malignant tumors (444 objects).
4) Wine (n = 178, d = 13, K = 3): This is a classication
problem with well-behaved class structures. There are
13 features, three classes, and 178 data vectors.
DAS et al.: AUTOMATIC CLUSTERING USING AN IMPROVED DE ALGORITHM 225
TABLE I
PARAMETERS FOR THE CLUSTERING ALGORITHMS
TABLE II
FINAL SOLUTION (MEAN AND STANDARD DEVIATION OVER 40 INDEPENDENT RUNS) AFTER EACH ALGORITHM
WAS TERMINATED AFTER RUNNING FOR 10
6
FEs WITH THE CS-MEASURE-BASED FITNESS FUNCTION
5) Vowel data set (n = 871, d = 3, K = 6): This data set
consists of 871 Indian Telugu vowel sounds. The data set
has three features, namely F
1
, F
2
, and F
3
, corresponding
to the rst, second and, third vowel frequencies, and six
overlapping classes {d (72 objects), a (89 objects), i (172
objects), u (151 objects), e (207 objects), o (180 objects)}.
B. Population Initialization
For the ACDE algorithm, we randomly initialize the ac-
tivation thresholds (control genes) within [0, 1]. The cluster
centroids are also randomly xed between X
max
and X
min
,
which denote the maximum and minimum numerical values of
any feature of the data set under test, respectively. For example,
in the case of the grayscale images (discussed in Section IV-F),
since the intensity value of each pixel serves as a feature, we
choose X
min
= 0 and X
max
= 255. To make the comparison
fair, the populations for both the ACDE and the classical
DE-based clustering algorithms (for all problems tested) were
initialized using the same random seeds. For the GCUK, each
string in the population initially encodes the centers of K
i
clusters, where K
i
= rand( ) K
max
. Here, K
max
is a soft
estimate of the upper bound of the number of clusters. The
K
i
centers encoded in the chromosome are randomly selected
points from the data set. In the case of the DCPSO algorithm,
the initial position of the ith particle

Z
i
(0) (for a binary
PSO) is xed depending on a user-specied probability P
ini
,
as follows:
Z
i,k
(0) =
_
0, if r
k
p
ini
1, if r
k
< p
ini
226 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART A: SYSTEMS AND HUMANS, VOL. 38, NO. 1, JANUARY 2008
TABLE III
MEAN CLASSIFICATION ERROR OVER NOMINAL PARTITION AND STANDARD DEVIATION OVER 40 INDEPENDENT RUNS, WHERE
EACH RUN WAS CONTINUED UP TO 10
6
FEs FOR THE FIRST FOUR EVOLUTIONARY ALGORITHMS (USING THE CS MEASURE)
TABLE IV
RESULTS OF THE UNPAIRED t-TEST BETWEEN THE BEST AND THE SECOND BEST PERFORMING ALGORITHMS
(FOR EACH DATA SET) BASED ON THE CS MEASURES OF TABLE II
TABLE V
MEAN AND STANDARD DEVIATIONS OF THE NUMBER OF FITNESS FEs (OVER 40 INDEPENDENT RUNS) REQUIRED
BY EACH ALGORITHM TO REACH A PREDEFINED CUTOFF VALUE OF THE CS VALIDITY INDEX
where r
k
is a uniformly distributed random number in [0, 1].
The initial velocity vector of each particle

V
i
(0) is randomly
set in the interval [5, 5] following [25].
C. Parameter Setup for the Compared Algorithms
We used the best possible parameter settings recommended
in [24] and [25] for the GCUK and DCPSO algorithms,
DAS et al.: AUTOMATIC CLUSTERING USING AN IMPROVED DE ALGORITHM 227
TABLE VI
MEAN CLASSIFICATION ERROR OVER NOMINAL PARTITION AND STANDARD DEVIATION OVER 40 INDEPENDENT
RUNS, WHICH WERE STOPPED AS SOON AS THEY REACHED THE PREDEFINED CUTOFF CS VALUE
TABLE VII
FINAL SOLUTION (MEAN AND STANDARD DEVIATION OVER 40 INDEPENDENT RUNS) WHEN EACH ALGORITHM
WAS TERMINATED AFTER RUNNING FOR 10
6
FEs WITH THE DB-MEASURE-BASED FITNESS FUNCTION
TABLE VIII
MEAN CLASSIFICATION ERROR OVER NOMINAL PARTITION AND STANDARD DEVIATION OVER 40 INDEPENDENT RUNS, WHERE EACH
RUN WAS CONTINUED UP TO 10
6
FEs FOR THE FIRST FOUR EVOLUTIONARY ALGORITHMS (USING THE DB MEASURE)
respectively. For the ACDE algorithm, we choose an optimal
set of parameters after experimenting with many possibilities.
Table I summarizes these settings. In Table I, Pop_size indicates
the size of the population, dim implies the dimension of each
chromosome, and P
ini
is a user-specied probability used for
initializing the position of a particle in the DCPSO algorithm.
For details on this issue, please refer to [25]. Once set, we allow
no hand tuning of the parameters to make the comparison fair
enough.
D. Simulation Strategy
In this paper, while comparing the performance of our ACDE
algorithm with other state-of-the-art clustering techniques, we
228 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART A: SYSTEMS AND HUMANS, VOL. 38, NO. 1, JANUARY 2008
TABLE IX
RESULTS OF THE UNPAIRED t-TEST BETWEEN THE BEST AND THE SECOND BEST PERFORMING
ALGORITHMS (FOR EACH DATA SET) BASED ON THE DB MEASURES OF TABLE VII
TABLE X
MEAN AND STANDARD DEVIATIONS OF THE NUMBER OF FITNESS FEs (OVER 40 INDEPENDENT RUNS) REQUIRED
BY EACH ALGORITHM TO REACH A PREDEFINED CUTOFF VALUE OF THE DB VALIDITY INDEX
TABLE XI
MEAN CLASSIFICATION ERROR OVER NOMINAL PARTITION AND STANDARD DEVIATION OVER 40 INDEPENDENT
RUNS, WHICH WERE STOPPED AS SOON AS THEY REACHED THE PREDEFINED CUTOFF DB VALUE
focus on three major issues: 1) quality of the solution as
determined by the CS and DB measures; 2) ability to nd the
optimal number of clusters; and 3) computational time required
to nd the solution.
For comparing the speed of the stochastic algorithms such
as GA, PSO, or DE, the rst thing we require is a fair time
measurement. The number of iterations or generations cannot
be accepted as a time measure since the algorithms perform
different amount of works in their inner loops, and they have
different population sizes. Hence, we choose the number of
tness function evaluations (FEs) as a measure of computation
time instead of generations or iterations.
Since four of the other algorithms used for comparison are
stochastic in nature, the results of two successive runs usually
DAS et al.: AUTOMATIC CLUSTERING USING AN IMPROVED DE ALGORITHM 229
Fig. 2. (a) Three-dimensional plot of the unlabeled iris data set using the rst three features. Clustering of iris data by (b) ACDE, (c) DCPSO, (d) GCUK,
(e) classical DE, and (f) average-link-based hierarchical clustering algorithm.
do not match for them. Hence, we have taken 40 independent
runs (with different seeds of the random number generator)
of each algorithm. The results have been stated in terms of
the mean values and standard deviations over the 40 runs in
each case. As the hierarchical agglomerative algorithm (marked
in Table II as average-link) used here does not use any
evolutionary technique, the number of FEs is not relevant to
this method. This algorithm is supplied with the correct number
of clusters for each problem, and we used the Ward updating
formula [58] to efciently recompute the cluster distances.
We used unpaired t-tests to compare the means of the results
produced by the best and the second best algorithms. The
unpaired t-test assumes that the data have been sampled from a
normally distributed population. From the concepts of the cen-
tral limit theorem, one may note that as sample sizes increase,
the sampling distribution of the mean approaches a normal
distribution regardless of the shape of the original population.
A sample size around 40 allows the normality assumptions
conducive for performing the unpaired t-tests [59].
The four evolutionary clustering algorithms can go with
any kind of clustering validity measure serving as their tness
functions. We executed two sets of experiments: one using the
CS-measure-based tness function that is shown in (20) while
the other using the DB-measure-based tness function that is
shown in (21), with all the four algorithms. For each data set,
the quality of the nal solution yielded by the four partitional
clustering algorithms has been compared with the average-
link metric-based hierarchical method in terms of the CS and
DB measures.
Finally, we would like to point out that all the algo-
rithms discussed here have been developed in a Visual C++
platform on a Pentium-IV 2.2-GHz PC, with a 512-kB
cache and a 2-GB main memory in Windows Server 2003
environment.
230 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART A: SYSTEMS AND HUMANS, VOL. 38, NO. 1, JANUARY 2008
E. Experimental Results
To judge the accuracy of the ACDE, DCPSO, GCUK, and
classical DE-based clustering algorithms, we let each of them
run for a very long time over every benchmark data set, until the
number of FEs exceeded 10
6
. Then, we note the nal tness
value, the number of clusters found, the intercluster distance,
i.e., the mean distance between the centroids of the clusters
(where the objective is to maximize the distance between
clusters), and the intracluster distance, i.e., the mean distance
between data vectors within a cluster (where the objective is to
minimize the intracluster distances). The latter two objectives
respectively correspond to crisp compact clusters that are well
separated. In the case of the hierarchical algorithm, the CS
value (as well as the DB index) has been calculated over the
nal results obtained after its termination. In columns 3, 4,
5, and 6 of Table II, we report the mean number of classes
found, the nal CS value, the intercluster distance, and the
intracluster distance obtained for each competitor algorithm,
respectively.
Since the benchmark data sets have their nominal partitions
known to the user, we also compute the mean number of
misclassied data points. This is the average number of objects
that were assigned to clusters other than according to the
nominal classication. Table III reports the corresponding mean
values and standard deviations over the runs obtained in each
case of Table II. Table IV shows results of unpaired t-tests
taken on the basis of the CS measure between the best two
algorithms (standard error of difference of the two means, 95%
condence interval of this difference, the t value, and the two-
tailed P value). For all the cases in Table IV, sample size = 40.
To compare the speeds of different algorithms, we selected
a threshhold value of CS measure for each of the data sets.
This cutoff CS value is somewhat larger than the minimum
CS value found by each algorithm in Table II. Now, we run
a clustering algorithm on each data set and stop as soon as
the algorithm achieves the proper number of clusters, as well
as the CS cutoff value. We then note down the number of
tness FEs that the algorithm takes to yield the cutoff CS value.
A lower number of FEs corresponds to a faster algorithm. In
columns 3, 4, 5, and 6 of Table V, we report the mean number
of FEs, the CS cutoff value, the mean and standard deviation
of the nal intercluster distance, and the mean and standard
deviation of the nal intracluster distance (on termination of
the algorithm) over 40 independent runs for each algorithm,
respectively. In Table VI, we report the misclassication errors
(with respect to the nominal classication) for the experiments
conducted for Table V. In this table, we exclude the hierar-
chical average-link algorithm as its time complexity cannot be
measured using the number of FEs. It is, however, noted that
the runtime of a standard hierarchical algorithm quadratically
scales [55].
Tables VIIXI exactly correspond to Tables IIVI with re-
spect to the experimental results, the only difference being
that all the experiments conducted for the former group of
tables use a DB-measure-based tness function [see (21)].
In all the tables, the best entries are marked in boldface.
Fig. 2 provides a visual feel of the performance of the four
Fig. 3. Dendrogram plot for the iris data set using the average-link hierar-
chical algorithm.
clustering methods over the iris data set. The data set has
been plotted in three dimensions using the rst three features
only (Fig. 3).
F. Discussion on the Results (for Real-Life Data Sets)
A scrutiny of Tables II and V reveals the fact that for the
iris data set, all the ve competitor algorithms terminated with
nearly comparable accuracy. The nal CS and DB measures
were the lowest for the ACDE algorithm. In addition, the ACDE
was successful in nding the nearly correct number of classes
(three for iris) over repeated runs. However, in Table III, we
also nd that the GCUK, DCPSO, and classical DE yield
two clusters, on average, for the iris data set. One of the clusters
corresponds to the Setosa class, whereas the others correspond
to the combination of Veriscolor and Virginica. This happens
because the latter two classes are considerably overlapping.
There are indexes other than the CS or DB measure available
in the literature, which yield two clusters for the iris data set
[60], [61]. Although the hierarchical algorithm was supplied
with the actual number of classes, its performance remained
poorer than all the four evolutionary partitional algorithms in
terms of the nal CS measure, the mean intracluster distance,
and the mean intercluster distance.
Substantial performance differences occur for the rest of the
more challenging clustering problems with a large number of
data items and clusters, as well as overlapping cluster shapes.
Tables II and V conform to the fact that the ACDE algorithm
remains clearly and consistently superior to the other three
competitors in terms of the clustering accuracy. For the breast
cancer data set, we observe that both the DCPSO and ACDE
yield very close nal values of the CS index, and both nd
two clusters in almost every run. Entries of Table IV testify
that the ACDE meets or beats its competitors in a statistically
signicant manner. We also note that the average-link-based
hierarchical algorithm remained the worst performer over these
data sets as well.
In Table VII, we nd that it is only in one case (for the
breast cancer data) that the classical DE-based algorithm yields
DAS et al.: AUTOMATIC CLUSTERING USING AN IMPROVED DE ALGORITHM 231
TABLE XII
PARAMETER SETUP OF THE CLUSTERING ALGORITHMS FOR THE IMAGE SEGMENTATION PROBLEMS
TABLE XIII
NUMBER OF CLASSES FOUND OVER FIVE REAL-LIFE GRAYSCALE IMAGES AND THE FOLIAGE IMAGE DATABASE USING THE
CS-BASED FITNESS FUNCTION (MEAN AND STANDARD DEVIATION OF THE NUMBER OF CLASSES FOUND OVER
40 INDEPENDENT RUNS, WHERE EACH RUN WAS CONTINUED FOR 10
6
FITNESS FEs)
TABLE XIV
AUTOMATIC CLUSTERING RESULT OVER FIVE REAL-LIFE GRAYSCALE IMAGES AND TWO IMAGE DATA SETS USING THE
CS-BASED FITNESS FUNCTION (MEAN AND STANDARD DEVIATION OF THE FINAL CS MEASURE FOUND OVER
40 INDEPENDENT RUNS, WHERE EACH RUN WAS CONTINUED FOR 10
6
FITNESS FEs)
TABLE XV
RESULTS OF THE UNPAIRED t-TEST BETWEEN THE BEST AND THE SECOND BEST PERFORMING
ALGORITHMS (FOR EACH DATA SET) BASED ON THE CS MEASURES OF TABLE XIV
232 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART A: SYSTEMS AND HUMANS, VOL. 38, NO. 1, JANUARY 2008
TABLE XVI
MEAN AND STANDARD DEVIATIONS OF THE NUMBER OF FITNESS FEs
(OVER 40 INDEPENDENT RUNS) REQUIRED BY EACH ALGORITHM TO
REACH A PREDEFINED CUTOFF VALUE OF THE CS VALIDITY
INDEX FOR THE IMAGE CLUSTERING APPLICATIONS
a lower DB measure, as compared to the ACDE. However, from
Table IX, we may note that this difference is not statistically
signicant.
Results of Tables III and VIII reveal that the ACDE yields
the least number of misclassied items once the clustering is
over. In this regard, we would like to mention that despite
the convincing performance of all the ve algorithms, none
of the experiments was without misclassication with respect
to the nominal classication, which was what we expected.
Interestingly, we found that the nal tness values obtained by
our evolutionary clustering algorithms were much better than
the tness of the nominal classication, which shows that the
misclassication could not be explained by the optimization
performance. Instead, misclassication is the result of the un-
derlying assumptions of the clustering tness criteria (such as
the spherical shape of the clusters), outliers in the data set,
errors in collecting data, and human errors in the nominal
solutions. This is indeed not a negative result. In fact, the
differences of a clustering solution based on statistical criteria
compared to the nominal classication can reveal interesting
data points and anomalies in the data set. In this way, a
clustering algorithm can be used as a very useful tool for data
preanalysis.
From Tables V and X, we can see that the ACDE was able
to reduce both the CS and DB index to the cutoff value within
the minimum number of FEs for majority of the cases. Both the
DCPSO and classical DE took lesser computational time than
the GCUK algorithm over most of the data sets. One possible
reason of this may be the use of less complicated variation
Fig. 4. (a) Original clouds image. (b) Segmentation by ACDE (K = 4).
(c) Segmentation by DCPSO (K = 4). (d) Segmentation with GCUK
(K = 4). (e) Segmentation with classical DE (provided K = 3).
operators (like mutation) in PSO and DE, as compared to the
operators used for GA.
V. APPLICATION TO IMAGE SEGMENTATION
A. Image Segmentation as a Clustering Problem
Image segmentation may be dened as the process of di-
viding an image into disjoint homogeneous regions. These
homogeneous regions usually contain similar objects of interest
or part of them. The extent of homogeneity of the segmented
regions can be measured using some image property (e.g.,
pixel intensity [11]). Segmentation forms a fundamental step
toward several complex computer vision and image analysis
applications, including digital mammography, remote sensing,
and land cover study. Segmentation of nontrivial images is
one of the most difcult tasks in image processing. Image
segmentation can be treated as a clustering problem, where
the features describing each pixel correspond to a pattern, and
each image region (i.e., segment) corresponds to a cluster [11].
Therefore, many clustering algorithms have widely been used
to solve the segmentation problem (e.g., K-means [62], fuzzy
C-means [63], ISODATA [64], Snob [65], and, recently, the
PSO- and DE-based clustering techniques [51], [53]).
DAS et al.: AUTOMATIC CLUSTERING USING AN IMPROVED DE ALGORITHM 233
Fig. 5. (a) Original robot image. (b) Segmentation by ACDE (K = 3).
(c) Segmentation by DCPSO (K = 2). (d) Segmentation with GCUK
(K = 3). (e) Segmentation with classical DE (provided K = 3).
B. Experimental Details and Results
In this section, we report the results of applying four evo-
lutionary partitional clustering algorithms (ACDE, DCPSO,
GCUK, and classical DE) to the segmentation of ve 256
256 grayscale images. The intensity level of each pixel serves
as a feature for the clustering process. Hence, although the
data points are single dimensional, the number of data items
is as high as 65 536. Finally, the same four algorithms have
been applied to classify an image database, which contains
28 small grayscale images of seven distinct kinds of foliages.
Each foliage occurs in the form of a 30 30 digital image. In
this case, each data item corresponds to one 30 30 image.
Taking the intensity of each pixel as a feature, the dimension of
each data point becomes 900. We run two sets of experiments
with two tness functions that are shown in (20) and (21).
However, to save space, we only report the CS-measure-based
results in this section. To tackle the high-dimensional data
points in the last aforementioned problem, we use a cosine dis-
tance measure that is described in (5) following the guidelines
in [11]. For the rest of the problems, the Euclidean distance
measure is used the same as before.
We carried out a thorough experiment with different parame-
ter settings of the clustering algorithms. In Table XII, we report
an optimal setup of the parameters that we found best suited
Fig. 6. (a) Original science magazine image. (b) Segmentation by ACDE
(K = 4). (c) Segmentation by DCPSO (K = 3). (d) Segmentation with
GCUK (K = 6). (e) Segmentation with classical DE (provided K = 3).
for the present image-related problems. With these sets of pa-
rameters, we observed each algorithm to achieve considerably
good solutions within an acceptable computational time. Note
that the parameter settings do not deviate much for the DCPSO
and GCUK algorithms than what is recommended in [24]
and [25].
Tables XIII and XIV summarize the experimental results
obtained over ve grayscale images in terms of the mean and
standard deviations of the number of classes found and the
nal CS measure reached at by the four adaptive clustering
algorithms. Table XV shows the results of the unpaired t-tests
taken based on the nal CS measure of Table XIV be-
tween the best two algorithms (standard error of difference
of the two means, 95% condence interval of this difference,
the t value, and the two-tailed P value). Table XVI records
the mean number of FEs required by each algorithm to reach
a predened cutoff CS value. This table helps in comparing
the speeds of different algorithms as applied to image pixel
classication.
Figs. 48 show the ve original images and their segmented
counterparts obtained using the ACDE, DCPSO, GCUK, and
classical DE-based clustering algorithms. Fig. 9 shows the
234 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART A: SYSTEMS AND HUMANS, VOL. 38, NO. 1, JANUARY 2008
Fig. 7. (a) Original peppers image. (b) Segmentation by ACDE (K = 7).
(c) Segmentation by DCPSO (K = 7). (d) Segmentation with GCUK
(K = 4). (e) Segmentation with classical DE (provided K = 8).
original foliage image database (unclassied). In Table XVII,
we report the best classication results achieved with this
database using the ACDE algorithm.
C. Discussion on Image Segmentation Results
From Tables XIIIXVI, one may see that our approach out-
performs the state-of-the-art DCPSO and GCUK over a variety
of image data sets in a statistically signicant manner. Not only
does the method nd the optimal number of clusters, but it also
manages to nd better clustering of the data points in terms
of the two major cluster validity indexes used in the literature.
From Table XVII, it is visible that the cluster number of the
proposed foliage image patterns is correctly determined by the
ACDE, and the cluster center images can represent common
and typical features of each class with respect to different types
of foliage.
The remote sensing image of Mumbai (a mega city of India)
in Fig. 8 bears special signicance in this context. Usually,
segmentation of such images helps in the land cover analysis of
different areas in a country. The newmethod yielded six clusters
for this image. A close inspection of Fig. 8(b) reveals that most
Fig. 8. (a) Original Indian Remote Sensing image of Mumbai. (b) Segmen-
tation by ACDE (K = 6). (c) Segmentation by DCPSO (K = 4). (d) Seg-
mentation with GCUK (K = 7). (e) Segmentation with classical DE (provided
K = 5).
Fig. 9. Nine hundred dimensional training patterns of seven different kinds of
foliages.
of the land cover categories have been correctly distinguished in
this image. For example, the Santa Cruz airport, the dockyard,
the bridge connecting Mumbai to New Mumbai, and many
other road structures have distinctly come out. In addition, the
predominance of one category of pixels in the southern part of
the image conforms to the ground truth; this part is known to
be heavily industrialized, and hence, the majority of the pixels
in this region should belong to the same class of concrete. The
Arabian Sea has come out as a combination of pixels of two
DAS et al.: AUTOMATIC CLUSTERING USING AN IMPROVED DE ALGORITHM 235
TABLE XVII
CLUSTERING RESULT OVER THE FOLIAGE IMAGE PATTERNS BY THE ACDE ALGORITHM
different classes. The seawater is found to be decomposed into
two classes, i.e., turbid water 1 and turbid water 2, based on the
difference of their reectance properties.
From the experimental results, we note that the ACDE
performs much better than the classical DE-based clustering
scheme. Since both algorithms use the same chromosome rep-
resentation scheme and start with the same initial population,
the difference in their performance must be due to the difference
in their internal operators and parameter values. From this, we
may infer that the adaptation schemes suggested for parameters
F and Cr of DE in (17) and (18) considerably improved
the performance of the algorithm at least for the clustering
problems covered here.
VI. CONCLUSION AND FUTURE DIRECTIONS
This paper has presented a new DE-based strategy for crisp
clustering of real-world data sets. An important feature of the
proposed technique is that it is able to automatically nd the
optimal number of clusters (i.e., the number of clusters does not
have to be known in advance) even for very high dimensional
data sets, where tracking of the number of clusters may be
well nigh impossible. The proposed ACDE algorithm is able to
outperform two other state-of-the-art clustering algorithms in a
statistically meaningful way over a majority of the benchmark
data sets discussed here. This certainly does not lead us to
claim that ACDE may outperform DCPSO or GCUK over
every data set since it is impossible to model all the possible
complexities of real-life data with the limited test suit that we
used for testing the algorithms. In addition, the performance
of DCPSO and GCUK may also be enhanced with a judicious
parameter tuning, which renders itself to further research with
these algorithms. However, the only conclusion we can draw at
this point is that DE with the suggested modications can serve
as an attractive alternative for dynamic clustering of completely
unknown data sets.
To further reduce the computational burden, we feel that
it will be more judicious to associate the automatic research
of the clusters with the choice of the most relevant features
compared to the process used. Often, we have a great number
of features (particularly for a high-dimensional data set like the
foliage images), which are not all relevant for a given operation.
Hence, future research may focus on integrating the automatic
feature-subset selection scheme with the ACDE algorithm.
The combined algorithm is expected to automatically project
the data to a low-dimensional feature subspace, determine the
number of clusters, and nd out the appropriate cluster centers
with the most relevant features at a faster pace.
REFERENCES
[1] I. E. Evangelou, D. G. Hadjimitsis, A. A. Lazakidou, and C. Clayton,
Data mining and knowledge discovery in complex image data using
articial neural networks, in Proc. Workshop Complex Reason. Geogr.
Data, Paphos, Cyprus, 2001.
[2] T. Lillesand and R. Keifer, Remote Sensing and Image Interpretation.
Hoboken, NJ: Wiley, 1994.
[3] H. C. Andrews, Introduction to Mathematical Techniques in Pattern
Recognition. New York: Wiley, 1972.
[4] M. R. Rao, Cluster analysis and mathematical programming, J. Amer.
Stat. Assoc., vol. 66, no. 335, pp. 622626, Sep. 1971.
[5] R. O. Duda and P. E. Hart, Pattern Classication and Scene Analysis.
Hoboken, NJ: Wiley, 1973.
[6] K. Fukunaga, Introduction to Statistical Pattern Recognition. New York:
Academic, 1990.
[7] B. S. Everitt, Cluster Analysis, 3rd ed. New York: Halsted, 1993.
[8] J. A. Hartigan, Clustering Algorithms. New York: Wiley, 1975.
[9] H. Frigui and R. Krishnapuram, A robust competitive clustering algo-
rithm with applications in computer vision, IEEE Trans. Pattern Anal.
Mach. Intell., vol. 21, no. 5, pp. 450465, May 1999.
[10] Y. Leung, J. Zhang, and Z. Xu, Clustering by scale-space ltering,
IEEE Trans. Pattern Anal. Mach. Intell., vol. 22, no. 12, pp. 13961410,
Dec. 2000.
[11] A. K. Jain, M. N. Murty, and P. J. Flynn, Data clustering: A review,
ACM Comput. Surv., vol. 31, no. 3, pp. 264323, Sep. 1999.
[12] E. W. Forgy, Cluster analysis of multivariate data: Efciency versus
interpretability of classication, Biometrics, vol. 21, no. 3, pp. 768769,
1965.
[13] C. T. Zahn, Graph-theoretical methods for detecting and describing
gestalt clusters, IEEE Trans. Comput., vol. C-20, no. 1, pp. 6886,
Jan. 1971.
[14] T. Mitchell, Machine Learning. New York: McGraw-Hill, 1997.
[15] J. Mao and A. K. Jain, Articial neural networks for feature extraction
and multivariate data projection, IEEE Trans. Neural Netw., vol. 6, no. 2,
pp. 296317, Mar. 1995.
[16] N. R. Pal, J. C. Bezdek, and E. C.-K. Tsao, Generalized clustering
networks and Kohonens self-organizing scheme, IEEE Trans. Neural
Netw., vol. 4, no. 4, pp. 549557, Jul. 1993.
[17] T. Kohonen, Self-Organizing Maps, vol. 30. Berlin, Germany: Springer-
Verlag, 1995.
[18] E. Falkenauer, Genetic Algorithms and Grouping Problems. Chichester,
U.K.: Wiley, 1998.
236 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART A: SYSTEMS AND HUMANS, VOL. 38, NO. 1, JANUARY 2008
[19] S. Paterlini and T. Minerva, Evolutionary approaches for cluster
analysis, in Soft Computing Applications, A. Bonarini, F. Masulli, and
G. Pasi, Eds. Berlin, Germany: Springer-Verlag, 2003, pp. 167178.
[20] J. H. Holland, Adaptation in Natural and Articial Systems. Ann Arbor,
MI: Univ. Michigan Press, 1975.
[21] S. Z. Selim and K. Alsultan, A simulated annealing algorithm for the
clustering problem, Pattern Recognit., vol. 24, no. 10, pp. 10031008,
1991.
[22] J. MacQueen, Some methods for classication and analysis of multivari-
ate observations, in Proc. 5th Berkeley Symp. Math. Stat. Probability,
1967, pp. 281297.
[23] R. Storn and K. Price, Differential evolutionA simple and efcient
heuristic for global optimization over continuous spaces, J. Glob. Optim.,
vol. 11, no. 4, pp. 341359, Dec. 1997.
[24] S. Bandyopadhyay and U. Maulik, Genetic clustering for automatic
evolution of clusters and application to image classication, Pattern
Recognit., vol. 35, no. 6, pp. 11971208, Jun. 2002.
[25] M. Omran, A. Salman, and A. Engelbrecht, Dynamic clustering using
particle swarm optimization with application in unsupervised image clas-
sication, in Proc. 5th World Enformatika Conf. (ICCI), Prague, Czech
Republic, 2005.
[26] J. Kennedy and R. Eberhart, Particle swarm optimization, in Proc. IEEE
Int. Conf. Neural Netw., 1995, pp. 19421948.
[27] A. Konar, Computational Intelligence: Principles, Techniques and Appli-
cations. Berlin, Germany: Springer-Verlag, 2005.
[28] P. Brucker, On the complexity of clustering problems, in Optimization
and Operations Research, vol. 157, M. Beckmenn and H. P. Kunzi, Eds.
Berlin, Germany: Springer-Verlag, 1978, pp. 4554.
[29] G. Hamerly and C. Elkan, Learning the K in K-means, in Proc. NIPS,
Dec. 813, 2003, pp. 281288.
[30] M. Halkidi and M. Vazirgiannis, Clustering validity assessment: Finding
the optimal partitioning of a data set, in Proc. IEEE ICDM, San Jose,
CA, 2001, pp. 187194.
[31] J. C. Dunn, Well separated clusters and optimal fuzzy partitions, J.
Cybern., vol. 4, pp. 95104, 1974.
[32] R. B. Calinski and J. Harabasz, A dendrite method for cluster analysis,
Commun. Stat., vol. 3, no. 1, pp. 127, 1974.
[33] D. L. Davies and D. W. Bouldin, A cluster separation measure, IEEE
Trans. Pattern Anal. Mach. Intell., vol. 1, no. 2, pp. 224227, Apr. 1979.
[34] M. K. Pakhira, S. Bandyopadhyay, and U. Maulik, Validity index for
crisp and fuzzy clusters, Pattern Recognit. Lett., vol. 37, no. 3, pp. 487
501, Mar. 2004.
[35] C. H. Chou, M. C. Su, and E. Lai, A new cluster validity measure and
its application to image compression, Pattern Anal. Appl., vol. 7, no. 2,
pp. 205220, Jul. 2004.
[36] V. V. Raghavan and K. Birchand, A clustering strategy based on a for-
malism of the reproductive process in a natural system, in Proc. 2nd Int.
Conf. Inf. Storage Retrieval, 1979, pp. 1022.
[37] C. A. Murthy and N. Chowdhury, In search of optimal clusters using
genetic algorithm, Pattern Recognit. Lett., vol. 17, no. 8, pp. 825832,
Jul. 1996.
[38] S. Bandyopadhyay, C. A. Murthy, and S. K. Pal, Pattern classication
with genetic algorithms, Pattern Recognit. Lett., vol. 16, no. 8, pp. 801
808, Aug. 1995.
[39] R. Srikanth, R. George, N. Warsi, D. Prabhu, F. E. Petri, and B. P. Buckles,
A variable-length genetic algorithm for clustering and classication,
Pattern Recognit. Lett., vol. 16, no. 8, pp. 789800, Aug. 1995.
[40] Y. C. Chiou and L. W. Lan, Theory and methodology genetic clustering
algorithms, Eur. J. Oper. Res., vol. 135, no. 2, pp. 413427, 2001.
[41] K. Krishna and M. N. Murty, Genetic K-means algorithm, IEEE Trans.
Syst., Man, Cybern., vol. 29, no. 3, pp. 433439, Jun. 1999.
[42] R. J. Kuo, J. L. Liao, and C. Tu, Integration of ART2 neural network
and genetic K-means algorithm for analyzing web browsing paths in
electronic commerce, Decis. Support Syst., vol. 40, no. 2, pp. 355374,
Aug. 2005.
[43] M. Halkidi, Y. Batistakis, and M. Vazirgiannis, On clustering vali-
dation techniques, J. Intell. Inf. Syst., vol. 17, no. 2/3, pp. 107145,
Dec. 2001.
[44] S. Theodoridis and K. Koutroubas, Pattern Recognition. New York:
Academic, 1999.
[45] C. Rosenberger and K. Chehdi, Unsupervised clustering method with
optimal estimation of the number of clusters: Application to image
segmentation, in Proc. IEEE ICPR, Barcelona, Spain, 2000, vol. 1,
pp. 656659.
[46] C.-Y. Lee and E. K. Antonsson, Self-adapting vertices for mask-layout
synthesis, in Proc. Model. Simul. Microsyst. Conf., M. Laudon and
B. Romanowicz, Eds., San Diego, CA, Mar. 2729, 2000, pp. 8386.
[47] H.-P. Schwefel, Evolution and Optimum Seeking, 1st ed. New York:
Wiley, 1995.
[48] L. J. Fogel, A. J. Owens, and M. J. Walsh, Articial Intelligence Through
Simulated Evolution. New York: Wiley, 1966.
[49] M. Sarkar, B. Yegnanarayana, and D. Khemani, A clustering algorithm
using an evolutionary programming-based approach, Pattern Recognit.
Lett., vol. 18, no. 10, pp. 975986, Oct. 1997.
[50] S. Paterlinia and T. Krink, Differential evolution and particle swarm
optimisation in partitional clustering, Comput. Stat. Data Anal., vol. 50,
no. 5, pp. 12201247, Mar. 2006.
[51] M. Omran, A. Engelbrecht, and A. Salman, Particle swarm optimiza-
tion method for image clustering, Int. J. Pattern Recognit. Artif. Intell.,
vol. 19, no. 3, pp. 297322, 2005.
[52] J. Kennedy and R. C. Eberhart, A discrete binary version of the parti-
cle swarm algorithm, in Proc. IEEE Conf. Syst., Man, Cybern., 1997,
pp. 41044108.
[53] M. Omran, A. P. Engelbrecht, and A. Salman, Differential evolution
methods for unsupervised image classication, in Proc. 7th CEC, 2005,
pp. 966973.
[54] S. Das, A. Konar, and U. K. Chakraborty, Two improved differential evo-
lution schemes for faster global search, in Proc. ACM-SIGEVO GECCO,
Washington, DC, 2005, pp. 991998.
[55] W. H. Day and H. Edelsbrunner, Efcient algorithms for agglomerative
hierarchical clustering methods, J. Classif., vol. 1, no. 1, pp. 124,
Dec. 1984.
[56] C. Blake, E. Keough, and C. J. Merz, UCI Repository of Machine Learn-
ing Database, 1998. [Online]. Available: http://www.ics.uci.edu/~mlearn/
MLrepository.html
[57] S. K. Pal and D. D. Majumder, Fuzzy sets and decision making ap-
proaches in vowel and speaker recognition, IEEE Trans. Syst., Man,
Cybern., vol. SMC-7, no. 8, pp. 625629, Aug. 1977.
[58] C. Olson, Parallel algorithms for hierarchical clustering, Parallel
Comput., vol. 21, no. 8, pp. 13131325, Aug. 1995.
[59] B. Flury, A First Course in Multivariate Statistics, vol. 28. Berlin,
Germany: Springer-Verlag, 1997.
[60] J. C. Bezdek and N. R. Pal, Some new indexes of cluster validity, IEEE
Trans. Syst., Man, Cybern., vol. 28, no. 3, pp. 301315, Jun. 1998.
[61] R. Kothari and D. Pitts, On nding the number of clusters, Pattern
Recognit. Lett., vol. 20, no. 4, pp. 405416, Apr. 1999.
[62] J. T. Tou and R. C. Gonzalez, Pattern Recognition Principles. London,
U.K.: Addison-Wesley, 1974.
[63] M. M. Trivedi and J. C. Bezdek, Low-level segmentation of aerial images
with fuzzy clustering, IEEE Trans. Syst., Man, Cybern., vol. SMC-16,
no. 4, pp. 589598, Jul. 1986.
[64] G. Ball and D. Hall, A clustering technique for summarizing multivariate
data, Behav. Sci., vol. 12, no. 2, pp. 153155, Mar. 1967.
[65] C. S. Wallace and D. M. Boulton, An information measure for
classication, Comput. J., vol. 11, no. 2, pp. 185194, Aug. 1968.
Swagatam Das was born in Kolkata, India, in 1980.
He received the B.E.Tel.E. and M.E.Tel.E. degrees in
control engineering in 2003 and 2005, respectively,
from Jadavpur University, Kolkata, where he is cur-
rently working toward the Ph.D. degree.
He is currently a Lecturer with the Department
of Electronics and Telecommunication Engineering,
Jadavpur University. He is the author or coauthor
of more than 25 papers published in international
journals and conference proceedings. His research
interests include evolutionary computing, swarm in-
telligence, pattern recognition, data mining, chaos theory, and bioinformatics.
Mr. Das has been a reviewer for several journals such as Pattern Recogni-
tion and IEEE/ACM TRANSACTIONS ON COMPUTATIONAL BIOLOGY AND
BIOINFORMATICS. He was the recipient of the Best Paper Award at the
Sixth International Conference on Intelligent Systems Design and Applications
(ISDA2006), Jinan, China.
DAS et al.: AUTOMATIC CLUSTERING USING AN IMPROVED DE ALGORITHM 237
Ajith Abraham (M96SM07) received the
Ph.D. degree from Monash University, Melbourne,
Australia, in 2001.
He is currently a Visiting Professor with the Cen-
tre for Quantiable Quality of Service in Com-
munication Systems (Q2S), Centre of Excellence,
Norwegian University of Science and Technol-
ogy (NTNU), Trondheim, Norway. Before joining
NTNU, he was working under the South Korean
Governments Institute of Information Technology
Advancement (IITA) Professorship Program at
Yonsei University, Seoul, Korea, and Chung-Ang University, Seoul. He was
a Visiting Researcher with Rovira i Virgili University, Tarragona, Spain, during
20052006 and is currently an Adjunct Professor with Jinan University, Jinan,
China, and Dalian Maritime University, Dalian, China. He has authored or
coauthored more than 300 research publications in peer-reviewed reputed
journals, book chapters, and conference proceedings. His primary research
interests are in computational intelligence, with a focus on using global
optimization techniques for designing intelligent systems. His application
areas include Web services, information security, Web intelligence, nancial
modeling, multicriteria decision making, data mining, etc. He is a regular
reviewer of IEEE Intelligent Systems, IEEE TRANSACTIONS ON KNOWLEDGE
AND DATA ENGINEERING, IEEE TRANSACTIONS ON NEURAL NETWORKS,
IEEE TRANSACTIONS ON FUZZY SYSTEMS, IEEE TRANSACTIONS ON
EVOLUTIONARY COMPUTATION, IEEE TRANSACTIONS ON SYSTEMS, MAN,
AND CYBERNETICS, and IEEE TRANSACTIONS ON POWER SYSTEMS.
Dr. Abraham serves on the Editorial Boards of more than a dozen interna-
tional journals and has also guest edited 23 special issues on various topics for
international journals. Since 2001, he has been actively involved in the Hybrid
Intelligent Systems and the Intelligent Systems Design and Applications series
of annual international conferences. He was the recipient of ve Best Paper
Awards.
Amit Konar (M97) received the B.E. degree from
the Bengal Engineering and Science University,
Shibpur, India, in 1983 and the M.E.Tel.E. and Ph.D.
degrees from Jadavpur University, Kolkata, India, in
1985 and 1994, respectively.
He is currently a Professor with the Department
of Electronics and Telecommunication Engineering,
Jadavpur University. He was a Visiting Professor for
the summer courses with the University of Missouri,
St. Louis, in 2006. His research areas include the
study of computational intelligence algorithms and
their applications to the entire domain of electrical engineering and computer
science. He has specically worked on fuzzy sets and logic, neurocomputing,
evolutionary algorithms, DempsterShafer theory, and Kalman ltering and has
applied the principles of computational intelligence in image understanding,
VLSI design, mobile robotics, and pattern recognition. He has supervised ten
Ph.D. theses. He currently serves on the Editorial Board of the International
Journal of Hybrid Intelligent Systems and the International Journal of Neuro-
computing. He is the author or coauthor of more than 130 papers published
in international journals and conference proceedings and is the author of ve
books, four of which have been published by Springer-Verlag, Germany, and
one has been published by CRC Press, Florida.
Dr. Konar was a recipient of the All India Council of Technical Education
(AICTE)-accredited 19972000 Career Award for Young Teachers for his
signicant contribution in teaching and research.