Automatic Unsupervised Data Classification Using Jaya Evolutionary Algorithm

Advanced Computational Intelligence: An International Journal (ACII), Vol.3, No.2, April 2016
DOI:10.5121/acii.2016.3204 35
AUTOMATIC UNSUPERVISED DATA CLASSIFICATION
USING JAYA EVOLUTIONARY ALGORITHM
Ramachandra Rao Kurada1
and Dr. Karteeka Pavan Kanadam2
1
Asst. Prof., Department of Computer Science & Engineering, Shri Vishnu Engineering
College for Women, Bhimavaram
2
Professor, Department of Information Technology, Shri Vishnu Engineering College for
Women, Bhimavaram
ABSTRACT
In this paper we attempt to solve an automatic clustering problem by optimizing multiple objectives such as
automatic k-determination and a set of cluster validity indices concurrently. The proposed automatic
clustering technique uses the most recent optimization algorithm Jaya as an underlying optimization
stratagem. This evolutionary technique always aims to attain global best solution rather than a local best
solution in larger datasets. The explorations and exploitations imposed on the proposed work results to
detect the number of automatic clusters, appropriate partitioning present in data sets and mere optimal
values towards CVIs frontiers. Twelve datasets of different intricacy are used to endorse the performance
of aimed algorithm. The experiments lay bare that the conjectural advantages of multi objective clustering
optimized with evolutionary approaches decipher into realistic and scalable performance paybacks.
KEYWORDS
Multi objective optimization, evolutionary clustering, automatic clustering, cluster validity indexes, Jaya
evolutionary algorithm.
1. INTRODUCTION
For the past three decades, majority of optimization problems demands improvement issues with
multiple objectives and are attracted towards evolutionary computation methodologies due their
simplicity of transformative calculation. The leverage of these evolutionary approaches are
flexible, to add, remove, modify any prerequisite regarding problem conceptualization, generation
of comparative Pareto set and has ability to tackle higher complexities than the mainstream
methods. These robust and powerful search procedures generally portray a set of candidate
solutions, selection procedure for mating, segmenting and re-assembling of set of several
solutions to produce new solutions. This is reflected by the speedily increasing of interest in the
field of evolutionary clustering with multi objective optimizations [1].
Data clustering is recognized as the most prominent unsupervised technique in machine learning.
This technique apportions a given dataset into homogeneous groups in view of some
likeness/disparity metric. Conventional clustering algorithms regularly make previous
assumptions about grouping a cluster structure, and adoptable with a suitable objective function
so that it can be optimized with classical or metaheuristic techniques. These estimations grade
inadequately when clustering presumptions are not hold in data [2].

36
The natural paradigm to fit the data distribution in the entire feature space, discovering exact
number of partitions is violated in single objective clustering algorithm if distinctive locales of
the component space contain clusters of diverged space. Estimating a combined solution which is
stable, confident and lower sensitivity to noise is unattainable by any single objective clustering
algorithm. Multi-objective clustering can be perceived as a distinct case of multi-objective
optimization, targeting to concurrently optimize several trade-off with numerous objectives under
specific limitations. The aim objective of multi-objective clustering is to disintegrate a dataset
into comparable groups, by exploiting the multiple objectives analogously [3-4].
In this paper, we provide an clustering algorithms underplayed with Jaya evolutionary algorithm
[15] to solve large set of objectives, for affricating factual automatic k determination, that are
interesting, suitable detachment prompted in data sets, and optimizing a set of cluster validity
indices (CVIs) simultaneously for encouraging most favourable convergence at final solutions.
For conquering high intra-cluster likeness and low inter-cluster likeness, this algorithm uses CVIs
as objective functions as mentioned in [5]. The set of internal and external validity indices used as
fitness functions in this paper are Rand, Adjusted Rand, Siloutte, Chou Be, Davies–Bouldin and
Xie–Beni indexs [6].
The remainder of this paper is organized as follows. Section II presents a review of recent
automatic clustering algorithms. In Section III, describes the scalability of the proposed
AutoJAYA algorithm and original Jaya evolutionary algorithm. The effectiveness of our scheme
is discussed in Section IV. Finally, Section V concludes this paper.
2. LITERATURE REVIEW
The survey published by Mukhopadhyay, Maulik and Bandyopadhyay, S. in 2015 argue the
importance of using multiobjective clustering in the domains of image segmentation,
bioinformatics, web mining with real time applications. The survey urges the importance of
Multiobjective clustering for optimizing multiple objective functions simultaneously. The authors
highlights the techniques for encoding, selection of objective functions, evolutionary operators,
schemes for maintaining non-dominated solutions and assorting an end solution [7].
In order to improve searching skills, in 2015, Abadi, & Rezaei combined of continuous ant
colony optimization and particle swarm optimization and proposed a strategy which is a
combination of these two algorithms with genetic algorithm, the results demonstrated were of
high capacity and resistance [8].
In 2015, Ozturk, Hancer and Karaboga used artificial bee colony algorithm in dynamic
(automatic) clustering discrete artificial bee colony as a similarity measure between the binary
vectors through Jaccard coefficient [9]. In 2014, Kumar and Chhabra, gravitational search
algorithm in real life problems, where prior information about the number of clusters is not
known in image segmentation domain to attain automatic segmentation of both gray scale and
colour images [10].
In 2014, Kuo, Huang, Lin, Wu and Zulvia determined the appropriate number of clusters and
assigns data points to correct clusters, with kernel function to increase clustering capability, in
this study they have used with bee colony optimization for attaining stable and accurate results
[11]. In 2014, Wikaisuksakul presented a multi-objective genetic algorithm for data clustering
methods, to handle the overlapping clusters with multiple objectives, using the fuzzy c-means
method. The real-coded values are encoded in a string to represent cluster centers and Pareto
solutions corresponding to a trade-off between the two objectives are finally produced [12].

37
In 2014, Mukhopadhyay, Maulik,Bandyopadhyay, and CoelloCoello, published two survey’s
with Part I and Part II on multiobjective evolutionary algorithms for data mining with
Evolutionary Computation [13-14]. Part I survey holds literature for basic concepts related to
multi-objective optimization in data mining and evolutionary approaches for feature selection and
classification. In part II the authors present the rules for association, clustering and other data
mining tasks related to different multi-objective evolutionary algorithms.
3. AUTOMATIC CLUSTERING ALGORITHM - AUTOJAYA
This paper attempts to constellate exact number of proper detachment in datasets automatically
without any human intervention during the algorithm execution. The objective functions for
assorting an end solution is postured as a multi-objective optimization problem, by optimizing a
customary of cluster validity indices concurrently. The proposed multi-objective clustering
technique uses a most recently developed evolutionary algorithm Jaya [15], based on multi-
objective optimization method as the underlying optimization strategy. The points are assigned
randomly to selected cluster centres based on Euclidean distance. The Rand, Adjusted Rand,
Silhouette, Chou Be, Davies–Bouldin and Xie–Beni CVIs are optimized simultaneously to
endorse the validity of aimed algorithm. Determinately, the aimed algorithm is able to perceive
both the best possible number of clusters and proper apportioning in the dataset. The efficiency of
the proposed algorithm is shown for twelve real-time data sets of varying complexities. The
results of this multi objective clustering techniques presented in Table 1, Table 2.
3.1. INITIALIZATION
To initialize the candidate solutions, the cluster centres are encoded as chromosomes. The
population α or number of candidate solutions are initialized randomly with n rows and m
columns. The set of solutions are represented as α୧,୨ሺ0ሻ = α୨
୫୧୬
+ randሺ1ሻ ∗ ൫α୨
୫ୟ୶
− α୨
୫୧୬
൯ and
each solution contains Max୩number of selected cluster centers, where Max୩ is randomly chosen
activation thresholds in [0, 1].
3.2. OBJECTIVE / FITNESS FUNCTIONS
A straightforward way to pose clustering as an optimization problem is to optimize some CVIs
that reflect the goodness of the clustering solutions. The correctness or accuracy of any
optimization method depends on its objective or fitness function being used in the algorithm [2-
3]. In this manner, it is regular to instantaneously advance with numerous of such measures for
optimizing distinctive attributes of data. To compute the distance between the centroid and
candidate solutions Euclidean distance measure is used, along with it the other objective functions
optimized simultaneously are the RI, ARI, DB, CS, XI, SIL CVIs [6].
.
3.3. JAYA EVOLUTIONARY ALGORITHM
Jaya is a simple, powerful optimization algorithm proposed by R Venkata Rao in 2015 for solving
the constrained and unconstrained optimization problems [15]. This algorithm is predicated on the
idea that the outcome obtained for a given problem should move towards the best solution and
evade the worst solution. This evolutionary approach does not require any particular algorithm-
specific control parameter, rather mandates common control parameters. The working procedure
of this evolutionary method is as follows:

38
Let fሺαሻ is the objective function to be minimized or maximized. At any iteration , assume that
there are ′m′ number of design variables i. e ሺj = 1,2, … mሻ, ′n′ number of candidate solutions (i.e.
population size, k = 1,2, … nሻ. Let the best candidate best obtains the best value of
fሺαሻሺi. e. fሺαሻୠୣୱ୲ሻ in the entire candidate solutions and the worst candidate worst obtains the
worst value of fሺαሻሺi. e. fሺαሻ୵୭୰ୱ୲ሻ in the entire candidate solutions. If α୨,୩,୧ is the value of the j୲୦
variable for the k୲୦
candidate during the i୲୦
iteration, then this value is modified as per the
following equation
α୨,୩,୧
′
= α୨,୩,୧ + rଵ,୨,୧උ൫α୨,ୠୣୱ୲,୧൯ − หα୨,୩,୧หඏ − rଶ,୨,୧උ൫α୨,୵୭୰ୱ୲,୧൯ − หα୨,୩,୧หඏ. ሺ1ሻ
whereα୨,ୠୣୱ୲,୧ is the value of the variable for the best candidate and α୨,୵୭୰ୱ୲,୧ is the value of the
variable j for the worst candidate. α୨,୩,୧
′
is updated value of α୨,୩,୧ and rଵ,୨,୧ and rଶ,୨,୧ are the two
random numbers for the j୲୦
variable during the i୲୦
iteration in the range [0,1]. The term
rଵ,୨,୧උ൫α୨,ୠୣୱ୲,୧൯ − หα୨,୩,୧หඏ indicates the tendency of the solution to move closer to the best solution
and the term rଶ,୨,୧උ൫α୨,୵୭୰ୱ୲,୧൯ − หα୨,୩,୧หඏ indicates the tendency of the solution to avoid the worst
solution. α୨,୩,୧
′
is accepted if it gives better function value. All the accepted function values at the
end of the termination are maintained and these values become the input to the next iteration. At
the end of each iteration all the accepted function values are retained and are fed as inputs to the
next iteration. This algorithm intends to reach best solution and tries to avoid worst solution.
The steps in Jaya algorithm are as follows:
1. Initialize population size, number of design variables and termination condition
2. Identify best and worst solution in the population
3. Modify the solutions based on best and worst solutions using (1)
4. Is the solution corresponding to α୨,୩,୧
ᇱ
better than the corresponding to α୨,୩,୧
a. accept and replace the previous solution
5. Else keep the previous solution
6. Is the termination criterion satisfied
a. report as optimum solution
7. Else go to Step 2
3.4. PROPOSED AUTOJAYA ALGORITHM
The working procedure of the aimed algorithm AutoJAYA is as follows:
1. Initialize the number of candidate solutions randomly as , in ݊ rows and m columns.
2. The set of solutions are represented as and
each solution contain number of selected cluster centers, where is randomly
chosen activation thresholds in [0, 1].
3. The fitness function to be maximized by default is Rand Index, and the solution
of at current generation with design variables is represented as
4. Spot the active cluster centers with value greater than 0.5 as best candidates,
solutions and less than 0.5 as worst candidates, solutions
5. For do
a. For each data vector , calculate its distance from all active cluster centers using
Euclidean distance
b. Assign to closest cluster

39
c. Evaluate each candidate solution quality using the fitness functions and find
solutions
d. Modify the solutions based on best and worst solutions using (1)
6. If the solution corresponding to better than the corresponding to accept and
replace the previous solution else keep the previous solution
4. EXPERIMENTAL ANALYSIS AND DISCUSSION
In this section, we report on experiments that use multi-objective clustering to identify partitions
in diverged set of datasets. The enactment of the aimed algorithm is pragmatic from the results
conquered by the following criteria elected, i.e. automatic k-detection, minimal consumption of
CPU time, low percentage of error rate and ideal values in CVIs.The number of iterations is
restricted to 30 independent runs for all the datasets. Table 1, Table 2 demonstrates the results of
AutoJAYA algorithm used over real-time datasets. These real-time datasets are extracted from
UCI Machine Learning Repository [18].The best results are shown as bold face.
Table 1. Results of Automatic Clustering algorithms Real-time datasets
Datasets (size*dim, k)
No. of
auto
clusters
CPU
time
(sec)
% of
error
rate
Mean value of CVIs
ARI RI HIM SIL CS DB
Iris (150*4, 3) 3.01 19.45 10.11 0.9815 0.9987 0.0922 0.9214 0.8416 0.7152
Wine (178*13, 3) 3.00 105.32 40.64 0.8414 0.7912 0.4910 0.6048 0.6417 0.8915
Glass (214*9, 6) 6.00 114.23 30.78 0.8000 0.9000 0.6018 0.6980 0.5297 1.0050
Ionosphere (351*3, 4) 2.00 30.12 8.01 0.9580 0.9877 0.9632 1.2580 1.0470 0.9784
Ecoil (336*7, 8) 8.00 45.12 11.48 0.9587 1.0145 0.9964 1.0258 0.9478 0.7859
Rocks (208*60, 2) 2.01 74.20 12.45 0.9971 0.9999 1.0000 1.0001 0.9478 0.9481
Parkinson (195*22, 2) 2.10 42.14 12.08 0.9240 0.9920 0.9974 0.9608 0.9814 1.0040
Diabetic (768*9, 2) 2.00 74.25 10.45 0.9871 0.9997 1.0024 0.9999 0.8478 0.8481
Segment (1500*20, 2) 3.01 1041.02 14.12 0.9631 0.9941 0.0786 0.9740 0.8994 0.4448
Weighting (500*8, 2) 5.99 110.29 15.23 1.0019 0.9663 0.0222 1.0025 0.9648 0.2560
Sonar (208*60, 2) 1.98 67.20 24.23 0.9988 1.0205 0.9635 0.9845 0.8458 0.8932
Rippleset (250*3, 2) 2.00 10.23 5.48 1.2800 1.0200 0.9874 1.0000 0.9990 0.9011
The AutoJAYA renders exact number of automatic clusters in Wine, Glass, Ionosphere, Ecoil,
Diabetic, Sonar, Rippleset, when compared to actual number of clusters (k) shown in column 1
of Table 1. The Rippleset dataset is the only dataset where AutoJAYA consumes minimum
amount of CPU time among all the datasets of varying size and complexity. In general, the CPU
time consumed by all the comparing datasets is between 5.48 sec to 1041.12 sec, which is purely
dependent on the volume and complexity of the dataset. Likewise, minimal percentage of error
rate is logged for the aimed algorithm in Iris and Wine datasets and the other likening datasets
registers the error rate between 5.48 % and 40.64%.
The CVIs in RI in Iris, DB in Wine, DB in Glass, CS in Ionosphere, RI in Ecoil and HIM in
Rocks and mines dataset registers optimal mean value, by endorsing the validity of the algorithm.
The CVIs DB in Parkinson, RI in Diabetic, RI in Segment, ARI in Weighting, RI in Sonar and
SIL Rippleset also follow the same tendency by submitting optimal mean value towards the
frontiers of CVIs. All these implications elevate the supremacy of proposed algorithm in
obtaining favourable results. The automatic clusters generated by AutoJAYA algorithm are
shown in Figure 1.

40
Figure 1. Automatic clusters produced by AutoJAYA in Glass and Weighting datasets
Table 2. Values of F-measure, ROC and SSE in real-time datasets using AutoJAYA algorithm
Datasets F-Measure ROC SSE
Iris 0.940 0.955 7.81
Wine 1.115 1.132 9.25
Glass 0.186 0.472 52.18
Ionosphere 0.501 0.485 726.10
Ecoil 0.479 0.464 695.10
Rocks 0.171 0.420 49.812
Parkinson 1.180 0.459 50.22
Diabetic 0.513 0.497 149.51
Segment 0.043 0.496 532.78
Weighting 0.893 0.941 520.9
Sonar 0.153 0.464 21.71
Rippleset 0.441 0.421 44.81
The results of F-measure, ROC area and Sum of Squared Error (SSE) of the proposed algorithm
on each real-time dataset are included in Table 2. The value deviations of F-measure, ROC area
and SSE amongst all the datasets is shown in Fig. 2. It is observed from both Table 1 and Figure 2
that the aimed algorithm has obtained better result in most of the cases for all the real-time
datasets.
Table 2 shows the corresponding values of F-Measure, ROC area and SSE for all comparing real-
time datasets. A significant remark on Table 2 is all the datasets tender better values for F-
measure and ROC area. The SSE value is very nominal for Iris and Wine datasets and relatively
mere optimal values for remaining datasets.
The culminating remarks after examining the applicability of AutoJAYA algorithm over real-time
is the aimed algorithm lodges better in most of the datasets in identifying the exact number of
automatic partitions, with minimum consumption of CPU and relatively low percentage of error
rate.. Hence these experiments speculate fact that AutoJAYA algorithm lay bare the advantages
of multiobjective clustering optimized with evolutionary approaches decipher into realistic and
scalable performance paybacks.

41
Figure 2. Values of F-Measure, ROC and SSE rendered by AutoJAYA in real-time datasets
5. CONCLUSIONS
In this article, a novel multi-objective clustering technique AutoJAYA based on the newly
developed Jaya Evolutionary algorithm is proposed. The explorations and exploitations enforced
on the technique, automatically determine the proper number of clusters, proper partitioning from
a given dataset and mere optimal values towards CVIs frontiers, by optimizing fitness functions
simultaneously.
Furthermore, it is observed that the aimed algorithm exhibits better performance in most of the
considered real-time datasets and is able to cluster appropriate partitions. Much further work is
needed to investigate the profound algorithm using different and more objectives, compare with
well established automatic clustering algorithm and to test the approach still more extensively
over diversified domains of engineering.
REFERENCES
[1] Zitzler, Eckart, Marco Laumanns, and Stefan Bleuler. "A tutorial on evolutionary multiobjective
optimization." Metaheuristics for multiobjectiveoptimisation. Springer Berlin Heidelberg, 2004. 3-37.
[2] SriparnaSaha, SanghamitraBandyopadhyay,"A new point symmetry based fuzzy genetic clustering
technique for automatic evolution of clusters",Information Sciences 179, 2009, pp. 3230–3246,
doi:10.1016/j.ins.2009.06.013
[3] SriparnaSaha, SanghamitraBandyopadhyay,"A symmetry based multiobjective clustering technique
for automatic evolution of clusters", Pattern Recognitions 43, 2010, pp. 738-751,
doi:10.1016/j.patcog.2009.07.004
[4] Eduardo Raul Hruschka, Ricardo J. G. B. Campello, Alex A. Freitas, and Andr´e C. Ponce Leon F. de
Carvalho, "A Survey of Evolutionary Algorithms for Clustering", IEEE transactions on systems, man,
and cybernetics—part c: applications and reviews, Vol. 39-2, 2009, pp. 133-155.
[5] NobukazuMatake, Tomoyuki Hiroyasu, Mitsunori Miki, TomoharuSenda, "Multiobjective Clustering
with Automatic k-determination for Large-scale Data", GECCO’07, July 7–11, 2007, London,
England, United Kingdom, ACM 978-1-59593-697-4/07/0007
[6] EréndiraRendón, Itzel Abundez, Alejandra Arizmendi and Elvia M. Quiroz., "Internal versus External
cluster validation indexes", International journal of computers and communications, 1(5), 2011.
[7] Mukhopadhyay, A., Maulik, U., &Bandyopadhyay, S. (2015). A Survey of Multiobjective
Evolutionary Clustering. ACM Computing Surveys (CSUR),47(4), 61.

42
[8] Abadi, M. F. H., &Rezaei, H. (2015). Data Clustering Using Hybridization Strategies of Continuous
Ant Colony Optimization, Particle Swarm Optimization and Genetic Algorithm. British Journal of
Mathematics & Computer Science, 6(4), 336.
[9] Ozturk, C., Hancer, E., &Karaboga, D. (2015). Dynamic clustering with improved binary artificial
bee colony algorithm. Applied Soft Computing, 28, 69-80.
[10] Kumar, V., Chhabra, J. K., & Kumar, D. (2014). Automatic cluster evolution using gravitational
search algorithm and its application on image segmentation. Engineering Applications of Artificial
Intelligence, 29, 93-103.
[11] Kuo, R. J., Huang, Y. D., Lin, C. C., Wu, Y. H., &Zulvia, F. E. (2014). Automatic kernel clustering
with bee colony optimization algorithm.Information Sciences, 283, 107-122.
[12] Wikaisuksakul, S. (2014). A multi-objective genetic algorithm with fuzzy c-means for automatic data
clustering. Applied Soft Computing, 24, 679-691.
[13] Mukhopadhyay, A., Maulik, U., Bandyopadhyay, S., &CoelloCoello, C. (2014). A survey of
multiobjective evolutionary algorithms for data mining: Part I. Evolutionary Computation, IEEE
Transactions on, 18(1), 4-19.
[14] Mukhopadhyay, A., Maulik, U., Bandyopadhyay, S., &Coello, C. (2014). Survey of multiobjective
evolutionary algorithms for data mining: Part II.Evolutionary Computation, IEEE Transactions on,
18(1), 20-35.
[15] R. Venkata Rao, "Jaya: A simple and new optimization algorithm for solving constrained and
unconstrained optimization problems",International Journal of Industrial Engineering Computations,
7, 2016, doi: 10.5267/j.ijiec.2015.8.004
[16] Ramachandra Rao Kurada, KanadamKarteekaPavan, AllamAppaRao,"Automatic Teaching–
Learning-Based Optimization-A Novel Clustering Method for Gene Functional
Enrichments",Computational Intelligence Techniques for Comparative Genomics, SpringerBriefs in
Applied Sciences and Technology.2015. 10.1007/978-981-287-338-5.
[17] Ramachandra Rao Kurada, KarteekaPavanKanadam, "A generalized automatic clustering algorithm
using improved TLBO framework", Int. Journal of Applied Sciences and Engineering Research, Vol.
4, Issue 4, 2015, ISSN 2277 – 9442.
[18] Lichman, M. (2013). UCI Machine Learning Repository [http://archive.ics.uci.edu/ml]. Irvine, CA:
University of California, School of Information and Computer Science.

Automatic Unsupervised Data Classification Using Jaya Evolutionary Algorithm

Recommended

More Related Content

What's hot (20)

Viewers also liked (17)

Similar to Automatic Unsupervised Data Classification Using Jaya Evolutionary Algorithm (20)

Recently uploaded (20)

Automatic Unsupervised Data Classification Using Jaya Evolutionary Algorithm