MRMR BA: A hybrid gene selection algorithm for cancer classification

Abstract

The microarray technology facilitates biologist in monitoring the activity of thousands of genes (features) in one experiment. This technology generates gene expression data, which are significantly applicable for cancer classification. However, gene expression data consider as high- dimensional data which consists of irrelevant, redundant, and noisy genes that are unnecessary from the classification point of view. Recently, researchers have tried to figure out the most informative genes that contribute to cancer classification using computational intelligence algorithms. In this paper, we propose a filter method (Minimum Redundancy Maximum Relevancy, MRMR) and a wrapper method (Bat algorithm, BA) for gene selection in microarray dataset. MRMR was used to find the most important genes from all genes in gene expression data, and BA was employed to find the most informative gene subset from the reduce set generated by MRMR that can contribute in identifying the cancers. The wrapper method using support vector machine (SVM) method with 10-fold cross-validation served as evaluator of the BA. In order to test the accuracy performance of the proposed method, extensive experiments were conducted. Three microarray datasets are used, which include: colon, Breast, and Ovarian. Same method procedure was performed to Genetic algorithm (GA) to conducts comparison with our proposed method (MRMR-BA). The results show that our proposed method is able to find the smallest gene subset with highest classification accuracy.

Full text

Journal of Theoretical and Applied Information Technology
30
th
June 2017. Vol.95. No 12
© 2005 – ongoing JATIT & LLS
ISSN:
1992-8645
www.jatit.org E-ISSN:
1817-3195
2610
MRMR BA: A HYBRID GENE SELECTION ALGORITHM
FOR CANCER CLASSIFICATION
1
OSAMA AHMAD ALOMARI,
1
AHAMAD TAJUDIN KHADER,
2
MOHAMMED AZMI AL-
BETAR,
1
LAITH MOHAMMAD ABUALIGAH
1
School of Computer Science, Universiti Sains Malaysia, Penang, Malayisa
2
Department of information technology, Al-Huson University College, Al-Balqa Applied University, Al-
Huson, Irbid-Jordan
Email:
1
oasa14_com004@student.usm.my,
1
tajudin@cs.usm.my,
2
mohbetar@bau.edu.jo,
1
Imqa15_com072@student.usm.my
ABSTRACT
The microarray technology facilitates biologist in monitoring the activity of thousands of genes (features)
in one experiment. This technology generates gene expression data, which are significantly applicable for
cancer classification. However, gene expression data consider as high- dimensional data which consists of
irrelevant, redundant, and noisy genes that are unnecessary from the classification point of view. Recently,
researchers have tried to figure out the most informative genes that contribute to cancer classification using
computational intelligence algorithms. In this paper, we propose a filter method (Minimum Redundancy
Maximum Relevancy, MRMR) and a wrapper method (Bat algorithm, BA) for gene selection in microarray
dataset. MRMR was used to find the most important genes from all genes in gene expression data, and BA
was employed to find the most informative gene subset from the reduce set generated by MRMR that can
contribute in identifying the cancers. The wrapper method using support vector machine (SVM) method
with 10-fold cross-validation served as evaluator of the BA. In order to test the accuracy performance of
the proposed method, extensive experiments were conducted. Three microarray datasets are used, which
include: colon, Breast, and Ovarian. Same method procedure was performed to Genetic algorithm (GA) to
conducts comparison with our proposed method (MRMR-BA). The results show that our proposed method
is able to find the smallest gene subset with highest classification accuracy.
Key words:
Bat-inspired algorithm, Cancer Classification, Gene Selection, MRMR, SVM.
1. INTRODUCTION
With the advent of DNA microarray technology,
the biologist has enabled to analysis thousands of
genes in one experiment. However, it is impossible
to examine the expression of these huge numbers
of genes through limited number of samples (high-
dimensional data) [1]. As detecting and classifying
of cancers are key issues in microarray technology,
the existing of this huge number of genes form a
challenge to classification algorithms. Gene
expression dataset is such kind of huge
dimensionality dataset extracted from DNA
microarray technology. DNA microarray
technology enables a new insight into the
mechanisms of living systems by possibility of
analyzing thousand of genes simultaneously and
gets significant information about cell’s function.
This particular information can be utilized for
diagnosis many diseases such as: Alzheimers
disease [2], diabetes diseases [3], and cancer
diseases [4].
Several studies have shown that most
genes measured in a DNA microarray experiment
are not relevant in the accurate classification of
different classes of the problem [5]. To avoid the
problem of the curse of dimensionality, gene
selection (which is commonly known as feature
selection) is process to find the most informative
genes with respect to improve predictive accuracy
of diseases [6]. Methods of gene selection are
divided into 3 categories: the wrapper approach,
the filter approach, and embedded approach. In the
first category, the classifier is employed to assess
the reliability of genes or genes subsets. In second
category, a filter method does not involve the
machine learning algorithm for removing
irrelevant and redundant features, instead uses the
principal characteristics of the training data to

Journal of Theoretical and Applied Information Technology
30
th
June 2017. Vol.95. No 12
© 2005 – ongoing JATIT & LLS
ISSN:
1992-8645
www.jatit.org E-ISSN:
1817-3195
2611
evaluate the significance of the genes subset or
genes [7]. Wrapper methods tend to give better
results but filter methods are usually
computationally less expensive than wrappers
[8].Embedded methods the last category of gene
selection approaches, which perform gene selection
in the process of training and are usually specific to
given classifier algorithm. Basically, it is
hybridization between filter and wrapper methods
to integrating the advantages of both approaches
that leads to finding informative genes with high
classification accuracy [9].
The meta-heuristic techniques have been
the mostly widely used in tackling gene selection
problems, and their performance has been proved
to be one of the better performing techniques that
have been used for solving gene selection problems
[10, 11]. Although many approaches have been
proposed to tackle gene selection problem; most of
them still suffer from problems of stagnation in
local optima and high computational cost, and it
cannot be guaranteed that the optimal subset of
genes will be acquired, these problems as a result
of huge search space [13][14]. Therefore, an
efficient gene selection algorithm is needed.
Bat algorithm is a metaheuristic
introduced by Xin-She Yang [15] is based on
echolocation of behavior of bats. The advantage of
bat algorithm is employ the idea of combining
population based algorithm and local search. This
combination result in providing the algorithm
capability of global diverse exploration and local
intensive exploitation which is the key point in
metaheuristic. Furthermore, It has been
successfully applied to a wide variety of
optimization problems such as Continuous
Optimization [15, 16], combinatorial optimization
and scheduling [17, 18], Inverse problem and
parameter estimation [16] and [19], Classifications
[20], Clustering [21], and Image processing [22].
Due to it is strength search capabilities of
BA algorithm; in this paper, we propose the
application of BA to select the informative genes
from microarray gene expression data. The new
adaptation of BA involve change its continuous
nature to binary nature. Moreover, when apply Bat
algorithm to huge dimensional data, it will be
encountered some challenging problem in
computational efficiency, similar to other
evolutionary algorithms. Toward avoiding these
problems and further improvement to the
performance of the Bat algorithm (BA), we
adopted a filtering method, minimum redundancy
maximum relevance (MRMR), as a preprocessing
step to reduce the dimensionality of microarray
datasets. Subsequently, Bat algorithm (BA) and
Support Vector Machine (SVM) will form as a
wrapper approach.
Bat algorithm produce a candidate gene
subsets from elite pool of genes generated by
MRMR, while SVM evaluate each candidate gene
subsets, we called it (MRMR-BA). Experiments
were carried out in order to evaluate the
performance of the proposed algorithm using
microarray benchmark datasets: Colon, Breast, and
Ovarian. Results of MRMR-BA were compared
with MRMR combined with GA (MRMR-GA)
algorithm. When tested using all benchmark
datasets, the MRMR-BA achieved better
performance in terms of highest classification
accuracy along with the lowest average number of
selected genes. This proves that BA algorithm
could be alternative approach for solving gene
selection problem.
The paper is organized as follow: Section
2 introduce the related work. Section3 defines gene
selection problem. Section 4 introduces a briefly
description of MRMR, Bat Algorithm (BA), Binary
Bat Algorithm (BBA) and SVM. Section 5
Illustrate the proposed method. The Experimental
Setup and results are presented in Section 6.
Finally, Section 7 conclude our paper and future
work.
2. RELATED WORK
In the recent years, extensive research has
been developed for gene selection problems.
Various techniques for gene selection have been
proposed. In the literature, there are several
algorithms for gene selection and cancer
classification using microarrays. Alshamlan et al.
[12] proposed a new hybrid gene selection method
namely Genetic Bee Colony (GBC) algorithm
where the hybridization was done by combining
Genetic Algorithm and Artificial Bee Colony
(ABC) algorithm. The GBC algorithm proves to
have a superior performance as it achieved the
highest classification accuracy along with the
lowest average number of genes. Shreem et al. [13]
proposed a new method that hybridizes the
Harmony Search Algorithm (HSA) and the Markov
Blanket (MB), called HSA- MB for gene selection
in classification problems. HSA utilizes naive
Baye’s classifier in its wrapper approach. During

Journal of Theoretical and Applied Information Technology
30
th
June 2017. Vol.95. No 12
© 2005 – ongoing JATIT & LLS
ISSN:
1992-8645
www.jatit.org E-ISSN:
1817-3195
2612
improvisation process, a new harmony solution is
passed to MB( i.e. the filter approach) for more im-
provement. Experiments were carried out on ten
microarray datasets. The HSA-MB performance
revealed comparable results to the state-of-art
approaches. El Akadi et al. [1] introduced a
framework consisting of a two-stage algorithm for
microarray data. In the first stage, Minimum
Redundancy Maximum Relevance (MRMR) is
employed to filter the genes. While in the second
stage, a GA is used to generate the gene subsets
and both classifiers Naïve Bayes (NB) and support
vector machine (SVM) are utilized for the
evaluation purpose.
3. GENE SELECTION PROBLEM
3.1 Problem Definition
Computationally, gene selection problems
can be expressed as a combinatorial optimization
problem in which the search space involves a set of
all possible subsets [23, 24]. This problem is
known to be an NP-hard problem [25], and is a
highly combinatorial search in nature. The number
of solutions in the search space exponentially
increases when the number of genes increases, and
there are [2N ] possible subsets of genes, where N
represents the number of genes.
3.2 Problem Formulation
The complete set of Genes is represented
by a binary string of length N, where a bit in the
string is set to 1 if it is to be kept, and set to 0 if it
is to be discarded, and N is the original number of
Genes. In context of optimization is called
solution representation. The problem formulation
is illustrated in Figure.1.
Figure 1: The problem formulation
4. RESEARCH BACKGROUND
4.1 MRMR
In this study, we employed minimum
redundancy maximum relevancy (MRMR) [26]
feature selection approach to address gene
selection problem. The MRMR tries to find the
most relevant features based on it correlation with
class label. Furthermore, minimize the
redundancy of the feature themselves. This
filtering process reveals with the features that it
has maximum relevancy and minimum
redundancy. To quantify both relevancy and
redundancy, mutual information (MI) is used to
estimate the mutual dependency of two variables.
MI is defined as in Eq.(1)) as follows:



(1)
Where X and Y are two features, P (X)
and P (Y ) are marginal probability functions, and
P (X, Y ) is the joint probability distribution. The
mutual information value for two completely
independently random variables is 0 [27].
Given fi , which represents the feature i,
and the class label c, the Maximum-Relevance
method selects the top m features in the descent
order of I(fi; c), i.e. the best m individual features
relevant to the class labels.







(2)
In order to eliminate the redundancy
among features, a Minimum-Redundancy criterion
is defined:






!


 (3)
A sequential incremental algorithm to
solve the simultaneous optimizations of
optimization criteria of Eqs. (2, and 3) is given as
the following. Suppose set G represents the set of
features and we already have Sm-1, the feature set
with m-1 genes, then the task is to select the m-th
feature from the set G-Sm -1. This feature is
selected by maximizing the single-variable
relevance minus redundancy function.
"
#$
%&'


(

)$


*

%&'

!

(4)
4.2 Bat Algorithm
Bat Algorithm (BA), introduce by Xin-
She Yang in 2010 [15], emulates the echolocation
behavior of bats. They are many kinds of bats in
nature. All of them have quiet similar behaviors
when navigating and hunting, whereas they are

Journal of Theoretical and Applied Information Technology
30
th
June 2017. Vol.95. No 12
© 2005 – ongoing JATIT & LLS
ISSN:
1992-8645
www.jatit.org E-ISSN:
1817-3195
2613
different in terms of size and weight. Among
them, microbats extensively used echolocation
feature. This feature assists microbats whiles it
seeks for prey and/or avoids obstacles in a
complete darkness. The behavior of microbats can
be formulated as novel optimization technique.
Furthermore, this behavior has been
mathematically modeled as follows:
In the BA, the artificial bat has position
vector, velocity vector, and frequency vector
which are updated during the course of iterations.
The BA can explore the search space through
position and velocity vectors (or updated
positions vectors). Each bat has a position Xi ,
frequency Fi , and Velocity Vi in a
d-dimensional search space.
The velocity, position, and frequency vectors is
updated in Eq.((5), (6), (7 ).
+

,-.+

,-

,(/012,3

(5)
4
5
6-.4
7
6-8
7
6-. (6)
Where Gbest is the best solution is
obtained so far and Fi represent the frequency of
the i-th bat which is update in each course of
iteration as follows:
9
7
9
:7;<
9
:=>
(9
:7;
? (7)
Where β is a random number of uniform
distribution in [0,1]. The BA employed a random
walk in order to improve its capability in
exploitation as given below.
4
;@A
4
BCD
-EF
G
(8)
Where E is a random number in [-1,1],
and A is the loudness of emitted sound. At each
iteration, the loudness and pulse emission I are
updated as follows:
H

,-. IH

, (9)
J

,-.J

K-L.(MN(O, (10)
where
α
and
γ
are constant parameters
lies between 0 and 1 used to update loudness rate
A
i
and pulse
rate (
r
i
). The pseudo code of the
algorithm is presented in Figure.2.
Figure 2: Pseudocode of bat algorithm.
4.3 Binary bat algorithm
In continuous version of BA, the artificial
bat can be moved around search space utilizing
positions and
velocity vectors (or updated
position vectors) within continuous real domain.
However, in dealing with
binary search space
where the position/ or solution is a series of 0’s
and 1’s binary bits. As binary search
space
dealing with only two numbers (”0” and ”1”),
the updating positions process cannot be
performed
using Eq.6. Therefore, a transfer
strategy should be found to reflect the velocity
vector value in changing
the elements of position
vector from ”0” to ”1” or vice versa.
Nakamura et al. [28] have introduced a
binary version of the Bat Algorithm restricting
the new bat’s
position to only binary values using
a sigmoid function as follows:
PQR
7S
T

<@
&UV
W
(11)
Therefore, Eq.11 can be replaced by:
4
7S
X.YPQR
7
S
TZ [
K\6]M^_`M (12)
In which [ a U(0,1). Therefore, Eq. (12)
can provide only binary values for each bat’s
coordinates in the boolean lattice, which stand for
the presence of absence of the features.
Initialize the bat population Xi (i=1,2,…..,n) and Vi
Define pulse frequency Fi
Initialize pulse rate ri and the loudness Ai
While (t < Max number of iterations)
Generate new solutions by adjusting frequency,
Updating velocities and positions [equations (5) to
(7)]
If (rand > ri )
Select a solution among the best solutions randomly
Generate a local solution around the selected best
solution
End if
Generate a new solution by flying randomly
If( rand < Ai & f(xi) < f(Gbest)
Accept the new solutions
Increase ri and reduce Ai
End if
Rank the bats and find the current Gbest
End while

Journal of Theoretical and Applied Information Technology
30
th
June 2017. Vol.95. No 12
© 2005 – ongoing JATIT & LLS
ISSN:
1992-8645
www.jatit.org E-ISSN:
1817-3195
2614
4.4 Support Vector Machine (SVM)
A support vector machine (SVMs) is a
supervised learning algorithm method used for
classification and regression [29]. SVM is a
powerful classification algorithm; due to it
efficient performance in pattern recognition
domain. For example, SVM classifier has
successfully applied to classify high-dimension
data, such as microarray gene expression data.
SVM constructs a hyperplanes or a set of
hyperplanes in a high- dimensional space, which
can be utilized for classification, regression, or
other tasks.
SVM has the capability to deal with
linear and nonlinear datasets. In linear data, SVM
tries to find an optimal separating hyperplane that
maximizes the margin between the training
examples and the class boundary. In
nonlinear data, we need to define a feature
mapping function X → φ(x). This mechanism that
defines feature mapping process is called kernel
function. There are three common functions:
• Polynomial kernel
bQc

c
!
TQc

dc
!
-eT
f
 (13)
• Radial basis kernel
bQc

c
!
T1
$ghi
*
$i
j
h
klm
n
, (14)
•
Sigmoidal kernel
bQc

c
!
T6]Ic

dc
!
(0 (15)
Where a and b are parameters define the
kernel’s behavior.
5. P
ROPOSED METHOD FOR GENE SELECTION
Based on aforementioned characteristic of
microarray dataset, it is impractical to adopt a
evolutionary algorithm such as BA algorithm
directly to such huge dimensionality data. A pre-
process should take place to overcome this
difficulty. Therefore, MRMR is employed to filter
noisy and redundant genes. It utilizes a series of
intuitive measures of relevance and redundancy to
pick out promising features for both continuous
and discrete datasets As shown in Eq.(1,2,3 and
4). The main role of MRMR is to select the genes
that have minimum redundancy for input genes
and maximum relevancy for cancer disease. To
further explore reduced gene subset and identify a
subset of informative genes, BA and SVM
classifier combined to seek for the better gene
subset in wrapper way, as shown in Figure.3.
BA has employed as a search technique
to figure out the near optimal gene subset from
the reduce set generated. Initially, as nature of
gene selection problem is binary. Therefore, BA
has changed to be applicable to binary problems,
as mentioned in section 3.3.
Based on BA pseducode (as shown in
Fig.1) BA algorithm starts to generate initial
population or group of bats. Each bat consists a
series of 0’s and 1’s bits, where bit value 1
represents that this gene is selected and bit value
0 represents that this gene is discarded. SVM has
utilized as a wrapper approach to evaluate each
candidate gene subset produced by BA. The
evaluation function is combined classification
accuracy and gene subset length, as shown in
following equation:
966M``opqor-?
s
s$t
(16)
Where α
q
R(D) is the average of
classification accuracy rate got by carrying out
ten multiple cross-validation with SVM
classifiers. In each round, SVM build a prediction
model on the training dataset based on gene
subset R and (class label)/Decision D, and a
testing will perform on the prediction model on
the testing dataset to obtain classification
accuracy.
|R| is the length of selected gene subset.
|C| is the total number of genes. u and β are two
parameters related to the significance of
classification quality and subset length. uϵ [0;1]
and β= (1-u). The classification accuracy is more
vital than subset length.
After initialization of bat population is
performed and each candidate solution is
evaluate, BA starts generate new solutions
according to equations (5,6,7). With a probability
range of pulse rate ri, each new bat location is
updated using a local search strategy around the
current selected best solutions. As already stated,
the probability of accepting the new solution over
the current solution depend on loudness Ai and if
the new solution is better the current solution. It is
obvious, that BA algorithm is controlled by two

Journal of Theoretical and Applied Information Technology
30
th
June 2017. Vol.95. No 12
© 2005 – ongoing JATIT & LLS
ISSN:
1992-8645
www.jatit.org E-ISSN:
1817-3195
2615
parameters: the pulse rate r
i
and the loudness rate
Ai. Typically, the rate of pulse emission ri
increases and the loudness Ai decreases when the
population become near to the local optimum.
6. E
XPERIMENTAL SETUP
&
RESULTS
The implementation of filter and wrapper
approaches was programmed using two
programming languages (i.e., java and matlab). In
filter approach, MRMR was implemented using
matlab whereas in wrapper approach, BA and
SVM were implemented using java.
The SVM used in this approach is based
on the one prepared in libsvm [30]. RBF kernel
was assign for SVM classifier. Furthermore, grid
search algorithm was running to tuning the
parameter of SVM classifier.
In this study, we tested the proposed
MRMR-BA method by comparing it with Genetic
algorithm. MRMR was carried out to filter-out the
genes. The highest 50 Genes based on MRMR
scores form a search space to BA
Figure 3: Framework of the Proposed Method For Gene Selection.
and GA. An SVM with 10-cross-validation is
applied to validate and assess each candidate gene
subset generated from both algorithms.
The 10-cross- validation method was
performed over each dataset. The dataset is
partitioned into (90%) for training set and
(10%)for testing set. Thus, SVM build a model
based on selected genes and test it on unseen data.
This process is repeated 10 times for statistical
validations. Both methods (BA and GA) are
evaluated base on two measurements: the
classification accuracy and the predictive gene
subset length.
6.1 Dataset
To evaluate the proposed MRMR-BA
approach, we carried out our experiments using 3
datasets of gene expression profile. These datasets
can be freely downloaded from
http://csse.szu.edu.cn/staff/zhuzx/Datasets.html.
The datasets and their characteristics are
summarized in Table 1.
Table 1: Dataset Information
Datasets
# Classes
# Samples
#
Genes
Colon
2 2000 62
Breast
2 24481 97
Ovarian
2 15154 253
6.2 Parameter Setting
We examine the effect of BA on three
microarray datasets with a different
dimensionality i.e., a large
dataset with 24481
genes (Breast), a medium dataset with 15154
genes (Ovarian) and a small dataset with
2000
genes (Colon). The parameter settings of BA for
Gene selection problem is listed in Table 2. The
parameters of the proposed algorithm were
selected based on our preliminary experiments.
They provide a good trade-off between solution
quality and the computational time needed to
reach good quality solutions.

Journal of Theoretical and Applied Information Technology
30
th
June 2017. Vol.95. No 12
© 2005 – ongoing JATIT & LLS
ISSN:
1992-8645
www.jatit.org E-ISSN:
1817-3195
2616
Our preliminary tests show that
increasing number of both iteration and
population has impact of the
algorithm
performance, but the computational time is
increased. We found the choosing of appropriate
values for pulse rate and loudness lead to generate
a good quality solution. Furthermore, the
parameter
values of GA are also determined as in
[31]. Table 2 also shows the parameter values for
the proposed
algorithm and GA.
6.3 Results and Discussion
In this research, we re-implement mRMR
with GA in order to conduct a fair comparison
compare with
mRMR-BA method. Moreover, we
set the stopping criterion to be maximum number
of iterations and we
executed 30 independent runs.
The average of both classification accuracy (ACC)
and gene
subset length (#G) were computed for
all runs for both MRMR-BA and MRMR-GA.
The results are presented
in Table 3.
From Table 3, the results show that the
performance of MRMR-BA is superior in terms of
average classification accuracy and average gene
subset size for all dataset comparing with MRMR-
GA. In respect to Breast dataset, MRMR-BA
outperforms MRMR-GA with higher classification
accuracy and lowest gene subset length. In
Ovarian dataset, MRMR-BA has achieved 100
classification accuracy with only 3.83 averages of
selected genes.
Table 2: parameter setting of BA and GA.
In contrast, the MRMR-GA obtained 100
classification accuracy with 19.35 average of
selected genes. For Colon dataset, the result of
MRMR-BA is quiet higher compared with
MRMR-GA, as the former obtained 93.12
classification accuracy with only 8.13 average of
selected genes. Whereas, MRMR-GA obtained
86.79 with 24.142 average of selected genes. This
positive results return to search characteristics of
BA algorithm in combining global diverse
exploration and local intensive exploitation that
result in boost it performance in selecting most
informative genes with respect to high
classification accuracy.
Table 3: Results of comparison between MRMR-BA and
MRMR-GA.
7. CONCLUSION
AND
FUTURE
WORK
In this paper, a new approach proposed
to solve gene selection problem which combined
MRMR, BA,
and SVM classifier. This approach is
a hybrid filter-wrapper approach. MRMR filter
approach
run in the beginning to figure out the
best genes. This process is to fine-tune the search
space. Then, the
reduced set of genes generated
from MRMR represent solution dimension for BA.
Each candidate gene subset
is evaluated by SVM
Classifier. Three cancer datasets were used to test
the performance of the proposed
approach.
Furthermore, a comparison with MRMR-GA
shows that our proposed approach achieved higher
classification accuracy with less number of
genes. The results exhibit that MRMR-BA is a
promising approach for solving gene selection
problem.
In the future work, experimental results
on more real and
benchmark datasets to verify
Algorithm
P
arameter
Ex
p
erime
n
t
al
Range
Selected
V
alue
BA
Number
of
iterations
50-100 100
Nu
m
b
er
of
artificial
bats
50-100 100
Fmin
0-1 0.3
Fmax
1-2 1
A
0-1 0.5
r
0-1 0.5
α
0-1 0.9
q
0-1 0.9
GA
Population
size - 50
Number of
generations
- 50
Crossover
probability
- 0.6
Mutation
rate - 0.5
Datasets
M
MRMR
-
BA
MRMR
-
GA
Breast
#G
18.3
23.86
A
C
C
88.6
86.606
Ovarian
#G
3.83
19.35
A
C
C
100
100
Colon
#G
8.13
24.142
A
C
C
93.12
86.79

Journal of Theoretical and Applied Information Technology
30
th
June 2017. Vol.95. No 12
© 2005 – ongoing JATIT & LLS
ISSN:
1992-8645
www.jatit.org E-ISSN:
1817-3195
2617
and extend this proposed algorithm. Moreover, an
enhancement to BA can
be done by hybridize
with existing local search algorithm to empower
exploitation process that result in
improve the
performance of BA.
REFERENCES:
[1] A. El Akadi, A. Amine, A. El Ouardighi, D.
Aboutajdine, A two-stage gene selection
scheme utilizing mrmr filter and ga wrapper,
Knowledge and Information Systems 26 (3)
(2011) 487–500.
[2] P. P. Panigrahi, T. R. Singh, Computational
studies on alzheimers disease associated
pathways and regulatory patterns using
microarray gene expression and network data:
Revealed association with aging and other
diseases, Journal of theoretical biology 334
(2013) 109–121.
[3] S. M. Yoo, J. H. Choi, S. Y. Lee, N. C. Yoo, J.
Choi, N. Yoo, Applications of dna microarray
in disease diagnostics.
[4] K.-H. Chen, K.-J. Wang, K.-M. Wang, M.-A.
Angelia, Applying particle swarm
optimization-based decision tree classifier for
cancer classification on gene expression data,
Applied Soft Computing 24 (2014) 773–780.
[5] T. R. Golub, D. K. Slonim, P. Tamayo, C.
Huard, M. Gaasenbeek, J. P. Mesirov, H.
Coller, M. L. Loh, J. R. Downing,
M. A. Caligiuri, et al., Molecular classification of
cancer: class discovery and class prediction by
gene expression monitoring, science 286 (5439)
(1999) 531–537.
[6] A. Jain, D. Zongker, Feature selection:
Evaluation, application, and small sample
performance, Pattern Analysis and Machine
Intelligence, IEEE Transactions on 19 (2)
(1997) 153–158.
[7] R. Kohavi, G. H. John, Wrappers for feature
subset selection, Artificial intelligence 97 (1)
(1997) 273–324.
[8] B.-Q. Li, L.-L. Hu, L. Chen, K.-Y. Feng, Y.-D.
Cai, K.-C. Chou, Prediction of protein domain
with mrmr feature selection and analysis, PLoS
One 7 (6) (2012) e39308.
[9] I. Guyon, A. Elisseeff, An introduction to
variable and feature selection, The Journal of
Machine Learning Research 3 (2003) 1157–
1182.
[10] P. Bermejo, J. A. G´amez, J. M. Puerta, A
grasp algorithm for fast hybrid (filter-wrapper)
feature subset selection in high-dimensional
datasets, Pattern Recognition Letters 32 (5)
(2011) 701–711.
[11] S. C. Yusta, Different metaheuristic strategies
to solve the feature selection problem, Pattern
Recognition Letters 30 (5) (2009) 525–534.
[12] H. M. Alshamlan, G. H. Badr, Y. A. Alohali,
Genetic bee colony (gbc) algorithm: a new gene
selection method for microarray cancer
classification, Computational biology and
chemistry 56 (2015) 49–60.
[13] S. S. Shreem, S. Abdullah, M. Z. A. Nazri,
Hybridising harmony search with a markov
blanket for gene selection problems,
Information Sciences 258 (2014) 108–121.
[14] B. Xue, M. Zhang, W. N. Browne, Particle
swarm optimization for feature selection in
classification: A multi-objective approach,
IEEE Transactions on Cybernetics 43 (6)
(2013) 1656–1671.
doi:10.1109/TSMCB.2012.2227469.
[15] X.-S. Yang, A new metaheuristic bat-inspired
algorithm, in: Nature inspired cooperative
strategies for optimization (NICSO 2010),
Springer, 2010, pp. 65–74.
[16] X.-S. Yang, A. Hossein Gandomi, Bat
algorithm: a novel approach for global
engineering optimization, Engineering
Computations 29 (5) (2012) 464–483.
[17] B. Ramesh, V. C. J. Mohan, V. V. Reddy,
Application of bat algorithm for combined
economic load and emission dispatch, Int. J. of
Electricl Engineering and Telecommunications
2 (1) (2013) 1–9.
[18] P. Musikapun, P. Pongcharoen, Solving multi-
stage multi-machine multi-product scheduling
problem using bat algorithm, in: 2nd
international conference on management and
artificial intelligence, Vol. 35, IACSIT Press
Singapore, 2012, pp. 98–102.
[19] J.-H. Lin, C.-W. Chou, C.-H. Yang, H.-L. Tsai,
et al., A chaotic levy flight bat algorithm for
parameter estimation in nonlinear dynamic
biological systems, source: Journal of
Computer and Information Technology 2 (2)
(2012) 56–63.
[20] S. Mishra, K. Shaw, D. Mishra, A new meta-
heuristic bat inspired classification approach for
microarray data, Procedia Technology 4 (2012)
802–806.

Journal of Theoretical and Applied Information Technology
30
th
June 2017. Vol.95. No 12
© 2005 – ongoing JATIT & LLS
ISSN:
1992-8645
www.jatit.org E-ISSN:
1817-3195
2618
[21] G. Komarasamy, A. Wahi, An optimized k-
means clustering technique using bat algorithm,
European Journal of Scientific Research 84 (2)
(2012) 26–273.
[22] S. Akhtar, A. Ahmad, E. Abdel-Rahman, A
metaheuristic bat-inspired algorithm for full
body human pose estimation, in: Computer and
Robot Vision (CRV), 2012 Ninth Conference
on, IEEE, 2012, pp. 369–375.
[23] B. Duval, J.-K. Hao, J. C. Hernandez
Hernandez, A memetic algorithm for gene
selection and molecular classification of cancer,
in: Proceedings of the 11th Annual conference
on Genetic and evolutionary computation,
ACM, 2009, pp. 201–208.
[24] M. Dash, H. Liu, Feature selection for
classification, Intelligent data analysis 1 (3)
(1997) 131–156.
[25] E. Amaldi, V. Kann, On the approximability of
minimizing nonzero variables or unsatisfied
relations in linear systems, Theoretical
Computer Science 209 (1) (1998) 237–260.
[26] Y. Cai, T. Huang, L. Hu, X. Shi, L. Xie, Y. Li,
Prediction of lysine ubiquitination with mrmr
feature selection and analysis, Amino acids 42
(4) (2012) 1387–1395.
[27] B. S¸en, M. Peker, A. C¸ avu¸so˘glu, F. V. C¸
elebi, A comparative study on classification of
sleep stage based on eeg signals using feature
selection and classification algorithms, Journal
of medical systems 38 (3) (2014) 1–21.
[28] R. Y. Nakamura, L. A. Pereira, K. Costa, D.
Rodrigues, J. P. Papa, X.-S. Yang, Bba: a
binary bat algorithm for feature selection, in:
Graphics, Patterns and Images (SIBGRAPI),
2012 25th SIBGRAPI Conference on, IEEE,
2012, pp. 291–297.
[29] V. N. Vapnik, An overview of statistical
learning theory, Neural Networks, IEEE
Transactions on 10 (5) (1999) 988–999. [30]
C.-C. Chang, C.-J. Lin, Libsvm: a library for
support vector machines, ACM Transactions on
Intelligent Systems and Technology (TIST) 2
(3) (2011) 27.
[31] Z. Zhu, Y.-S. Ong, M. Dash, Markov blanket-
embedded genetic algorithm for gene selection,
Pattern Recognition 40 (11) (2007) 3236–3248.