See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/330832032

Dragonfly Algorithm: Theory, Literature Review, and Application in Feature

Selection

Chapter  in  Studies in Computational Intelligence · January 2020

DOI: 10.1007/978-3-030-12127-3_4

CITATIONS READS
50 455

5 authors, including:

Majdi Mafarja Ali Asghar Heidari

Birzeit University National University of Singapore
86 PUBLICATIONS   4,776 CITATIONS    130 PUBLICATIONS   6,750 CITATIONS   

SEE PROFILE SEE PROFILE

Hossam Faris Seyedali Mirjalili

University of Jordan Griffith University
212 PUBLICATIONS   8,522 CITATIONS    339 PUBLICATIONS   34,887 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

CPyEFECT: Comprehensive Python E-mail Features Extraction and Classification Tool View project

Whale Optimization Algorithm (WOA): theories, variants, and applications View project

All content following this page was uploaded by Ali Asghar Heidari on 29 August 2021.

The user has requested enhancement of the downloaded file.

Dragonfly Algorithm: Theory, Literature
Review, and Application in Feature
Selection

Majdi Mafarja, Ali Asghar Heidari, Hossam Faris, Seyedali Mirjalili, and
Ibrahim Aljarah

Abstract In this chapter, a wrapper-based feature selection algorithm is de-

signed and substantiated based on the binary variant of Dragonfly Algorithm
(BDA). DA is a successful well-established metaheuristic that revealed supe-
rior efficacy in dealing with various optimization problems including feature
selection. In this chapter we are going to test the performance of this algo-
rithm with a special type of datasets, that contain a huge number of features
with low number of samples. This type of datasets makes the optimization
process harder, because of the large search space, and the lack of adequate
samples to train the model. The experimental results showed the ability of
DA to deal with this type of datasets better than other optimizers in the
literature. Moreover, an extensive literature review for the DA is provided in
this chapter.

M. Mafarja
Department of Computer Science, Faculty of Engineering and Technology, Birzeit
University, PoBox 14, Birzeit, Palestine
e-mail: [email protected]
A. A. Heidari
School of Surveying and Geospatial Engineering, University of Tehran, Tehran, Iran
e-mail: as [email protected]
H. Faris and I. Aljarah
King Abdullah II School for Information Technology, The University of Jordan, Am-
man, Jordan
e-mail: {hossam.faris,i.aljarah}@ju.edu.jo
S. Mirjalili
Institute of Integrated and Intelligent Systems, Griffith University, Nathan, Brisbane,
QLD 4111, Australia
e-mail: [email protected]

1
2 Mafarja et. al

1 Introduction

Some creatures’ behaviors have been the inspiration source for many suc-
cessful optimization algorithms. The main behavior that inspired many re-
searchers to develop new algorithms was the strategy that those creatures
use to seek the food sources. Ant Colony Optimization (ACO) [21, 22] and
Artificial Bees Colony (ABC) [49] were originally inspired by the behavior of
ants and bees respectively in locating food sources and collecting their food.
Swarming behavior is another source of inspiration that was used to propose
new optimization algorithms. Particle Swarm Optimization (PSO) [23, 36]
is a primary swarm based optimization algorithm that mimics the swarming
behavior of birds.
The key issue about all the previously mentioned creatures is that they
live in groups or folks (called swarms) [51]. An individual in those swarms
usually makes a decision based on local information from itself and from the
interactions with the other swarm members, also from the environment. Such
interactions are the main reason that contribute to the improvement of the
social intelligence in these swarms. Most of the swarms contain different or-
ganisms from the same species (bees, ants, birds, etc.). By the intelligent col-
laboration (or swarm intelligence (SI)) between all individuals of the swarm,
they have been found to be successful in carrying out specific tasks [30, 91].
Nature-inspired algorithms are population-based metaheuristics algorithms,
that tempt to manipulate a set of solutions in each generation of the optimiza-
tion process [37, 38, 42, 51]. Recently, many such algorithms were proposed
and proved their ability to solve different complex optimization problems
such as global function optimization [11, 39, 40, 41, 43], clustering analysis
[3, 7, 9, 86], spam and intrusion detection systems [8, 10, 26, 28], optimiz-
ing neural networks [5, 6, 25, 27], link prediction [15], and software effort
estimation [31].
As Feature Selection is known to be an NP-hard problem [4], metaheuristic
algorithms in general and specially the population-based algorithms showed
a superior performance in solving this problem.
PSO has been widely used to tackle FS problems. An improved PSO with
local search strategy for FS was proposed in [77]. Another PSO approach with
two crossover mechanisms was presented in [19]. (author?) [79] proposed
a hybrid particle swarm optimization with shuffled frog leaping algorithm
feature selection approach. An improved ACO was also proposed in [50]. A
hybrid FS approach that combines differential evolution (DE) with ABC was
proposed in [98].
Grey Wolf Optimizer (GWO) [76] is a recent SI algorithm that mimics the
hierarchical organization of the grey wolves in nature. The GWO has been
widely used in FS methods with much success [56]. Antlion Optimizer (ALO)
was proposed by (author?) [71] in 2015, FS was one of the fields where ALO
was applied successfully [67]. Whale Optimization Algorithm (WOA), is an-
other SI algorithm that was recently proposed by (author?) [75]. WOA was
DA: Theory, Literature Review, and Application in Feature Selection 3

successfully applied in many FS methods [69, 70]. Grasshopper Optimization

Algorithm (GOA) [84] is another example of the nature inspired algorithms
that was recently proposed and applied to many optimization problems in-
cluding FS [4, 47, 66]. More recently, a Salp Swarm Algorithm (SSA) was
proposed by (author?) [73] and was used as a FS selection mechanism in
[1].
Dragonfly Algorithm (DA) was recently proposed by (author?) [72]. DA
is inspired by the behavior of dragonflies, and proved its ability outperform
other well-regarded optimizers in the literature. The DA has shown an excel-
lent performance for several continuous, discrete, single-objective and multi-
objective optimization problems compared to several state-of-the-art meta-
heuristic and evolutionary algorithms such as PSO and DE. In this chapter,
the binary version of DA is used to serve as a selection mechanism and applied
to medical datasets of high dimensional with small number of samples.

2 Feature Selection Problem

With the emergence of the high dimensional data in almost all of the real life
fields, the knowledge discovery (KDD) process becomes very complex [59].
The high dimensionality of datasets usually causes many problems for the
data mining techniques (e.g., classification), like over-fitting, and hence, de-
creasing the model performance, and increasing the required computational
time. Therefore, reducing data dimensionality becomes highly demanded [57].
FS is a preprocessing step that aims to enhance the performance of the data
mining and machine learning techniques by eliminating the redundant irrev-
erent features.
In the FS process, two major steps should be carefully considered when
designing a new FS method; selection (search) and evaluation [59]. In the se-
lection step, the feature space has to be searched for a feature subset with a
minimal number of features, and reveals the highest model performance (e.g.,
classification accuracy when considering a classification technique). In the
selection step, three different mechanisms can be considered; complete, ran-
dom, and heuristic search mechanisms. When dealing with high-dimensional
datasets, the complete search strategies (i.e., brute force) are impractical
since 2N feature subsets have to be generated and evaluated for a dataset
with N features. The random mechanisms are also impractical to be used
with FS methods since the search process is controlled by random factors
that may lead the search process to be as worst as the complete search, in
the worst case. Heuristic search algorithms are the most suitable mechanisms
to be used with FS methods since the search process is guided by a heuristic
that navigates the process towards finding the near-optimal solutions.
The evaluation criterion is another aspect that should be carefully consid-
ered when designing a FS method. In this regard, FS methods can be divided
4 Mafarja et. al

into three main types; filters, wrappers, and embedded methods. In filter ap-
proaches [61, 62, 63, 64, 65], feature subsets are evaluated depending on the
relations between features themselves, and their correlation with the class
attribute without considering any particular classifier [60]. Thus, a feature
with high score is selected to build the classification model in a further step.
In the wrapper based methods [1, 29, 70], each feature subset is evaluated
based on the performance of a particular classifier. Embedded methods (e.g.,
regularization) [17, 55, 59] learn which features enhance the performance of
the model (e.g., accuracy), while the model is being created.
In summary, filter methods are considered as the fastest FS methods,
since no external tools are required to be employed; nevertheless the model
performance is uncertain. Wrappers are slower than filters, while they usu-
ally obtain higher accuracy. Embedded methods come in between filters and
wrappers in terms of the model performance and the computational time.

3 Dragonfly Algorithm

Dragonfly Algorithm (DA) is a recently well-established population-based

optimizer proposed by Mirjalili in 2016 [72]. The DA was developed based on
the hunting and migration strategies of dragonflies. The hunting technique is
known as static swarm (feeding), in which all members of a swarm can fly in
small clusters over a limited space for discovering food sources. The migration
strategy of dragonflies is called dynamic swarm (migratory). In this phase,
the dragonflies are willing to soar in larger clusters, and as a result, the
swarm can migrate. Dynamic and static groups are shown in Figs. 1 and
2. Likewise other swarm-based methods, the operators of DA perform two
main concepts: diversification, motivated by the static swarming activities,
and intensification, encouraged by the dynamic swarming activities.
In DA, five types of behaviors are designed as follows. In next parts, X
is the position vector, Xj is the j-th neighbor of the X, and N denotes the
neighborhood size [81]:

• Separation is a strategy that dragonflies use to separate themselves from

other agents. This procedure is formulated as Eq. (1):
N
X
Si = − X − Xi (1)
j=1

• Alignment shows how an agent will set its velocity with respect to the
velocity vector of other adjacent dragonflies. This concept is modeled based
on Eq. (2):
PN
j=1 Vj
Ai = (2)
N
DA: Theory, Literature Review, and Application in Feature Selection 5

C1
DP
EBC P
A1

D1
1 1
1E OD1 D C
O
DP A B
C A B
E C P P

DP
1

A
A1
1E

PB
B
O 11
D
C
1
P
1

A
D
B1 P

O
B
AC
D
O
B
C

E BC 1 PP
A1 1
1 P
O 11 1E P D
O C
D P
1

A C A B

A1 E 1
1
PB

A1
DP
BC
1
A
D

O
1
P
O
B
C

P
O1

O
B
A1

A
P
1

C
1E OD1
O1
O
G
A C
2
HHG FEF
PB I I1 1 1 D E
GF J O1
HH G F J 1A BC D
1
HHG GFF E 1
I IJ1 1O
E GFF I I1 1 1 D E A11 1 1 C
DE HHG JA OC1 GGFBF
A
A
11
B
B
1C

11
1 I
J 1 1 D I J1 1O
C J 1A B1C
E
D
DE
1 D
1C
J 1 B
G A11 1 1 C
HHG FEF B
1
D H H E
I I1 1 1 D E
J O
J 1A BC11 D
B
A 11
B
11 I I1 1 1 D E A11 1 1 C
JA OC1 B
J1 B 1 D
A11 1 1 C
B

A
Fig. 1: Dynamic swarming behaviors (each geometric object shows a special
type of agents)

D1
D1 P
E
A1
DP
BC CA11 P
1
1E O 11
D D D
O C
O C
A C
PB A B

A1
A B P

P
1
E
1
DP
A1
O 1
1
BC
1
B1 P
1

D
P

AC
D
O
B
C

DP
A
1E

PB
D
C D A
O
B
PP C1
AC B1
1
E BC
D
P

C
A1 P 1
D
O C
1
O 11
1
DP A B
1E D E B
A C A1 O1 A1
P
PB 1
O 11
1 A
1E

PB
D
C O A
D1 1

A
B

D1
1
B A1
C G
HHG GFF C1 G
I IJ1 1O
J 1A BC
E
1D
1 D
E
1
BI HIJAHGBOFCEDF E I HIJAHGBOFCEDF E
1 1 1
1
1 1 1

A1 A11
B
11
1C
HHG GFF
I IJ1 1O
E
B
J1 1
A1 1 1 C
D J1 1
A11 1 1 C
D
1 D1
E G 1 B B
J 1A BC
A11
B
1 1
1 D
1C
HHG FEF
J O
D
I I1 1 1 D E O
C1
D J 1A BC11 D
P
A C A11 1 1 C
B
O
O
1
P
1

2
B
G
O A HHG FEF
B
I
A B
I 1 1 1 DE
J O
J 1A BC11 D
A11 1 1 C
B

C
B
Fig. 2: Static swarming behaviors.
6 Mafarja et. al

where Vj indicates the velocity vector of the j-th neighbor.

• Cohesion indicates the inclination of members to move in the direction of
the nearby center of mass. This step is formulated as in Eq. (3):
PN
j=1 xj
Ci = −X (3)
N
• Attraction shows to the propensity of members to move towards the food
source. The attraction tendency among the food source and the ith agent
is performed based on Eq. (4):

Fi = Floc − X (4)

where Floc is the food source’s location.

• Distraction shows the proclivity of dragonflies to keep themselves away

from an conflicting enemy. The distraction among the enemy and the ith
dragonfly is performed according to Eq. (5):

Ei = Eloc + X (5)

where Eloc is the enemy’s location.

In DA, the fitness of food source and position vectors are updated based
on the fittest agent found so far. Moreover, the fitness values and positions
of the enemy are computed based on the worst dragonfly. This fact can
assist DA in converging towards more promising regions of the solution
space and in turn, avoiding from non-promising areas. The position vectors
of dragonflies are updated based on two rules: the step vector (∆X) and
the position vector. The step vector indicates the dragonflies’ direction of
motion and it is calculated as in Eq. (6):

∆Xt+1 = (sSi + aAi + cCi + f Fi + eEi ) + wXt (6)

where s, w, a, c, f , and e show the weighting vectors of different compo-

nents.
The location vector of members is calculated as in Eq. (7):

Xt+1 = Xt + ∆Xt+1 (7)

where t is iteration.
Pseudocode of DA is shown in Algorithm 1.
DA: Theory, Literature Review, and Application in Feature Selection 7

Algorithm 1 Pseudocode of DA
Initialize the swarm Xi (i = 1, 2, . . . , n)
Initialize ∆Xi (i = 1, 2, . . . , n)
while (end condition is not met) do
Evaluate all dragonflies
Update (F ) and (E)
Update coefficients (i., e., w, s, a, c, f, and e)
Attain S, A, C, F , and E (based on Eqs. (1 to 5))
Update step vectors (∆Xt+1 ) by Eq. (6)
Update Xt+1 by Eq. (7)
Return the best agent

4 Literature Review

In this section, previous researches on DA and their core findings are reviewed
in detail. From the time of proposing DA, several works have focused on the
efficacy of DA or tried to improve its efficacy on tasks such as photovoltaic
systems [80], extension of RFID network lifetime [44], economic emission dis-
patch (EPD) problems [16]. (author?) [34] used the multi-objective version
of DA to solve the design problem of a nearly zero energy building (nZEB),
along with six other evolutionary methods. In [82], a hybrid DA approach
with extreme learning machine (ELM) was proposed. In this system, DA was
used to optimize the number of nodes and their associated weights in the
hidden layer of the ELM. (author?) [78] proposed a DA based approach to
solve the economic load dispatch (ELD) problem with valve point effect. DA
was used in [20] to estimate the location of the unknown wireless nodes that
are randomly deployed in a designated area. The binary version of DA was
used with a multi-class SVM classifier within an FS method [24]. (author?)
[46] employed DA to optimize the firing angle and the size of the thyristor
controlled series capacitor (TCSC). In [83], a self-adaptive DA was used to
tackle the multilevel segmentation problem.
In 2017, a memory-based hybrid DA with particle swarm optimization
(PSO) has been developed to deal with global optimization [53]. Song et
al. [87] developed an enhanced DA with elite opposition learning (EOL) to
tackle global optimization tasks. In [2], DA was used to solve the 0-1 knap-
sack problems. Another application that was solved using DA is the static
economic dispatch incorporating solar energy as in [90]. DA was used as a FS
search strategy in addition to tune the parameters of ELM in [97]. A multi
objective version of DA (MODA) was used as an adaptive engine calibration
algorithm to control the engine parameters [33]. (author?) [68] proposed
a FS method that employed the binary version of DA (BDA) as a search
strategy. In [92, 93], DA was used as a tool to optimize SVM parameters.
In software engineering field, DA was also applied. In [89], DA was used as
an optimization tool to select the most important test cases that satisfy all
8 Mafarja et. al

requirements. An improved version of DA based on elite opposition learning

strategy was used to solve some global optimization problems [87].
(author?) [90] proposed the use of DA in optimizing the generation costs
of all the participating units in a solar thermal economic dispatch system.
Memory based Hybrid DA (MHDA) was proposed in [53] to solve numerical
optimization problems. DA was used in [32] to optimize the performance of
the Pull-in Voltage performance in a MEMS switch. A hybrid model that
combined DA with PSO was proposed in [85] to solve the optimal power
flow (OPF) problem. DA was also used to select the optimal cluster heads in
Radio Frequency Identification (RFID) networks [45]. (author?) [48] used
the DA to optimize the orthotropic infinite plates with a quasi-triangular
cut-out.
In [96], DA was utilized as an efficient optimizer for turn-mill operations.
(author?) [35] proposed an enhanced binary DA to deal with infant cry clas-
sification scenarios. (author?) [52] proposed a hybrid DA with pattern search
for developing and studying of tilt integral derivative controllers. (author?)
[54] developed an improved fractional DA to be integrated with cluster cloud
models. In [95], a modified cumulative DA in cooperation with neural net-
works was proposed. They also proposed a map-reduce cluster architecture
to predict the performance of students using the DA-based approach. In [94],
DA was employed to optimize the size and cost of static var compensator for
enhancement of voltage profiles in power systems. (author?) [13] proposed
a new hybrid DA with support vector regression (SVR) for voltage stability
analysis in power systems. (author?) [18] proposed a DA-based load bal-
ancing approach to allocate resources in cloud computing. (author?) [14]
applied DA to synthesis of antenna arrays.
To sum up, in the reviewed papers, experimental results show the efficient
performance of DA in terms of balancing the exploration and exploitation,
speed of convergence, and satisfactory local optima (LO) avoidance.

5 Binary DA (BDA)

The DA algorithm was originally designed to solve the continuous optimiza-

tion problems. Thus, using DA to tackle the FS problem requires changing the
structure of the algorithm to deal with binary solutions. Transfer functions
can be employed to convert the continuous solution to a binary one (au-
thor?) [74], by generating a probability of changing a position’s elements
to 0 or 1 based on the value of the step vector (velocity) of the ith search
agent in the dth dimension in the current iteration (t) as an input param-
eter. (author?) [72] employed the transfer function in Eq. (8) to calculate
the probability of changing the continuous positions to binary. This trans-
fer function belongs the V-shaped (see Fig. 3) transfer functions category as
denoted by (author?) [74].
DA: Theory, Literature Review, and Application in Feature Selection 9

|vdi (t)|
T (vdi (t)) = p (8)
1 + (vdi (t))2
The result T (vki (t)), obtained from Eq. (8) is then used to convert the i-th
element of the position vector to 0 or 1 according to Eq. 9

¬Xt r < T (vki (t))


X(t + 1) = (9)
Xt r ≥ T (vki (t))
where r is a random number in the [0,1] interval.

0.9

0.8

0.7

0.6
Probability

0.5

0.4

0.3

0.2

0.1

0
-4 -3 -2 -1 0 1 2 3 4
x

Fig. 3: V-Shaped Transfer function [74]

5.1 BDA-based wrapper feature selection

As the FS is a binary optimization problem, a binary vector of length N

(where N is the number of features in the dataset), with 1 for the selected
feature and 0 for the non selected feature, is used in the proposed approach
(see Fig. 4). The KNN classifier [12] evaluates the selected feature subsets.
Here, KNN classifier was used for its simplicity which makes it easy to imple-
ment, also it is a nonparametric classifier and hence can be used with different
datasets. Moreover, KNN has been widely used with many FS methods and
demonstrated to have competitive performance with a wide range of real life
datasets [58].
In wrapper FS methods, two contradictory objectives should be consid-
ered; classification accuracy and reduction rate. The fitness function Eq. (10)
considers both aspects. α and β are two parameters corresponding to the im-
10 Mafarja et. al

F1 F2 F3 F4 ... FN-2 FN-1 FN

1 0 1 1 ... 0 1 0

Fig. 4: Solution Representation

portance of classification quality and subset length, α is in the [0,1] interval

and β =(1 −α) is adapted from [66].

|R|
F (X) = α ∗ γ(X) + β ∗ (1 − ) (10)
|N |
where γ(X) represents the classification accuracy while using X feature sub-
set, |R| is the number of selected features and |N | is the total number of
features in the original dataset. Therefore, α (a number in the interval [0,1])
and β (1-α) [66] parameters were used to represent the importance of classi-
fication quality and the reduction rate, respectively.
Pseudocode of BDA-based feature selection technique is shown in Algo-
rithm 2.

Algorithm 2 Pseudocode of BDA-based feature selection algorithm

Initialize the candidate features Xi (i = 1, 2, . . . , n)
Initialize ∆Xi (i = 1, 2, . . . , n)
while (end condition is not met) do
Evaluate all dragonflies based on Eq. (10)
Update (F ) and (E)
Update coefficients (i., e., w, s, a, c, f, and e)
Attain S, A, C, F , and E (based on Eqs. (1 to 5))
Update step vectors (∆Xt+1 ) by Eq. (6)
Calculate T (∆X) using Eq. (8)
Update Xt+1 by Eq. (9)
Return the best set of features

6 Results and discussions

To test the proposed approach, 9 high dimensional datasets with low num-
ber of samples were used. As can be seen form Table 1, nine datasets have
226 distinct categories, 50308 samples (patients) and 2308-15009 variables
(genes). All datasets are accessible in a public source1 . A train/test model is
1
http://www.gems-system.org/
DA: Theory, Literature Review, and Application in Feature Selection 11

utilized to validate the performance of the proposed BDA approach, where

80% of the data were used for training purposes, while the remaining 20%
were used for testing [70]. All approaches were implemented using MATLAB
2017a and the results were repeated for 30 independent times on an Intel
Core i5 machine, 2.2 GHz CPU and 4 GB of RAM. Table 2 presents the
detailed parameter settings for utilized methods.

Table 1: Details of high-dimensional small instances datasets [88]

Dataset name No. of Samples No. of Features No. of Classes

11 Tumors 174 12533 11
14 Tumors 308 15009 26
Brain Tumor1 90 5920 5
Brain Tumor2 50 10367 4
Leukemia1 72 5327 3
Leukemia2 72 11225 3
SRBCT 83 2308 4
Prostate Tumor 102 10509 2
DLBCL 77 5469 2

Table 2: The parameter settings

Parameter Value
Population size 10
Number of iterations 100
Dimension Number of features
Number of runs for each technique 30
α in fitness function 0.99
β in fitness function 0.01
a in bGWO [2 0]
Qmin Frequency minimum in BBA 0
Qmax Frequency maximum in BBA 2
A Loudness in BBA 0.5
r Pulse rate in BBA 0.5
G0 in BGSA 100
α in BGSA 20

6.1 Results and analysis

In order to assess the performance of BDA algorithm on the high dimensional

datasets, we tested six well-know metaheuristics algorithms for comparison
purposes. All algorithms were implemented and ran on the same environment
12 Mafarja et. al

in order to make a fair comparison. For all algorithms, the average classifica-
tion accuracy, average selected features, and average fitness values for the 30
independent runs are reported. Note that the best results in the subsequent
results are highlighted in boldface. Moreover, a Wilcoxon signed-rank test
is executed to state if there is a significant difference between the reported
results with the significance interval 95% (α = 0.05).
Table 3 shows the average classification accuracy over the 30 runs of dif-
ferent algorithms. As can be seen from Table 3, the proposed BDA shows
the best performance compared to all other approaches. BGSA comes in the
second place by obtaining the best results in three datasets, and then bGWO
outperformed other approaches in only one dataset. Comparing BDA with
BGSA; the second best approach, it can be obviously seen that BDA records
the best results in 6 datasets, while BGSA is outperforming other approaches
in 2 datasets, and both approaches obtained the same results in dealing with
Brain Tumor1 dataset. According to Table 4, which presents the p-values of
BDA versus other approaches, there is a significant difference in the perfor-
mance of the proposed approach and the other peers. These results verify
that the proposed BDA maintains a more stable balance between diversifica-
tion and intensification, and this leads to higher classification rates compared
to other peers.

Table 3: Average classification accuracy for all approaches

Dataset bGWO BGSA BBA BGA BPSO BDA

11 Tumors 0.858 0.819 0.656 0.616 0.758 0.879
14 Tumors 0.576 0.481 0.556 0.428 0.477 0.724
Brain Tumor1 0.944 0.944 0.894 0.848 0.726 0.889
Brain Tumor2 0.720 0.800 0.543 0.745 0.416 0.857
DLBCL 0.754 0.875 0.848 0.868 0.818 0.875
Leukemia1 0.869 0.658 0.756 0.761 0.663 0.933
Leukemia2 0.867 0.867 0.922 0.821 0.818 0.933
Prostate Tumor 0.878 0.943 0.921 0.834 0.806 0.902
SRBCT 0.882 0.884 0.859 0.760 0.746 0.941
DA: Theory, Literature Review, and Application in Feature Selection 13

Table 4: p-values between BDA and other approaches for the classification
accuracy

BDA vs.
Dataset
bGWO BGSA BBA BGA BPSO
11 Tumors 1.61E-05 1.86E-11 1.20E-11 1.27E-11 1.19E-11
14 Tumors 2.54E-11 1.53E-11 2.64E-11 2.52E-11 2.84E-11
Brain Tumor1 1.69E-14 1.69E-14 1.70E-01 7.24E-13 9.13E-13
Brain Tumor2 1.03E-10 1.43E-06 8.50E-12 2.53E-11 8.71E-12
DLBCL 4.16E-14 NaN 2.98E-04 1.75E-02 9.21E-13
Leukemia1 1.17E-13 1.57E-13 3.37E-13 9.38E-13 1.02E-12
Leukemia2 1.69E-14 1.69E-14 2.14E-02 7.65E-13 5.57E-13
Prostate Tumor 1.19E-04 2.98E-09 4.36E-03 9.49E-12 2.88E-12
SRBCT 1.69E-14 1.17E-13 3.48E-11 8.41E-13 8.09E-13

Inspecting the results in Table 5, it can be obviously seen that BDA selects
the minimal number of features on 55% of the datasets. It is followed by BBA
which came in the second place by outperforming other approaches in 45% of
the datasets. we see that no other approach is capable to compete with those
two approaches. P-values in Table 6 also show that BDA can significantly
outperform other approaches (namely bGWO, BGSA, BGA, and BPSO) in
all datasets, while on five datasets out of nine, it can significantly outperform
BBA approach.

Table 5: Average selected features for all approaches

Dataset bGWO BGSA BBA BGA BPSO BDA

11 Tumors 6233.23 6258.00 5082.73 6261.87 6261.50 5660.17
14 Tumors 8511.53 7564.97 6207.97 7500.77 7489.47 7206.77
Brain Tumor1 2926.63 2896.13 2389.60 2960.17 2950.57 2269.77
Brain Tumor2 5141.93 5106.47 4173.83 5156.87 5196.80 4285.23
DLBCL 2691.43 2699.57 2208.83 2731.20 2733.30 1884.90
Leukemia1 2620.63 2662.03 2077.73 2660.40 2664.53 1850.47
Leukemia2 5530.73 5523.27 4472.83 5613.60 5605.80 4466.07
Prostate Tumor 5391.43 5242.90 4124.50 5269.43 5247.00 4372.77
SRBCT 1126.73 1144.80 925.37 1161.10 1148.97 909.57
14 Mafarja et. al

Table 6: p-values between BDA and other approaches for the number of
selected features

BDA vs.
Dataset
bGWO BGSA BBA BGA BPSO
11 Tumors 3.02E-11 3.02E-11 3.52E-07 3.01E-11 3.02E-11
14 Tumors 3.46E-10 1.46E-10 6.68E-11 2.22E-09 4.36E-09
Brain Tumor1 3.02E-11 3.00E-11 3.37E-01 3.02E-11 3.01E-11
Brain Tumor2 3.01E-11 3.01E-11 1.02E-01 3.01E-11 3.01E-11
DLBCL 3.01E-11 3.00E-11 2.60E-08 3.01E-11 3.02E-11
Leukemia1 3.00E-11 3.01E-11 7.20E-05 3.01E-11 3.01E-11
Leukemia2 3.01E-11 3.01E-11 8.36E-01 3.02E-11 3.01E-11
Prostate Tumor 3.02E-11 3.01E-11 2.87E-02 3.01E-11 3.02E-11
SRBCT 3.00E-11 3.00E-11 6.41E-01 2.99E-11 3.00E-11

As mentioned before, the adopted fitness function considers two objectives

of the wrapper FS approach, i.e., the classification accuracy and the number
of selected features. Observing the results in Tables 7 and 8, it can be seen the
BDA significantly outperforms bGWO, BGSA, BGA, and BPSO approaches
on all datasets, and on 55% of the datasets when compared to BBA. In
addition to the superior performance of BDA approach, it recorded the fastest
convergence rates among other approaches in most of the datasets as can be
seen in Fig. 5, which shows the convergence speed of bGWO, BGSA, BBA,
and BDA over the iterations. The reasons for satisfying convergence curves
in majority of datasets is that BDA has a high capacity for performing the
focused intensification tendencies in last steps and extra diversification jumps
in initial phases.
DA: Theory, Literature Review, and Application in Feature Selection 15

Table 7: Average fitness values for all approaches

Dataset bGWO BGSA BBA BGA BPSO BDA

11 Tumors 0.145 0.184 0.321 0.757 0.777 0.124
14 Tumors 0.425 0.519 0.417 0.473 0.504 0.278
Brain Tumor1 0.060 0.060 0.080 0.837 0.876 0.114
Brain Tumor2 0.282 0.203 0.393 0.600 0.653 0.146
DLBCL 0.248 0.129 0.125 0.843 0.845 0.127
Leukemia1 0.135 0.344 0.212 0.862 0.875 0.069
Leukemia2 0.137 0.137 0.069 0.863 0.884 0.070
Prostate Tumor 0.126 0.062 0.024 0.846 0.859 0.102
SRBCT 0.121 0.119 0.105 0.864 0.889 0.062

Table 8: p-values between BDA and other approaches for the fitness values

BDA vs.
Dataset
bGWO BGSA BBA BGA BPSO
11 Tumors 7.67E-09 2.99E-11 3.01E-11 2.91E-11 2.94E-11
14 Tumors 3.01E-11 3.00E-11 3.01E-11 2.98E-11 2.96E-11
Brain Tumor1 3.01E-11 3.00E-11 1.59E-07 2.82E-11 2.54E-11
Brain Tumor2 2.94E-11 2.95E-11 3.01E-11 2.83E-11 2.68E-11
DLBCL 2.92E-11 2.95E-11 1.20E-01 2.57E-11 2.77E-11
Leukemia1 2.95E-11 3.01E-11 3.00E-11 2.76E-11 2.69E-11
Leukemia2 2.91E-11 2.91E-11 1.67E-06 2.80E-11 2.63E-11
Prostate Tumor 1.28E-09 9.79E-05 4.97E-11 2.75E-11 2.80E-11
SRBCT 2.98E-11 2.96E-11 2.77E-05 2.79E-11 2.80E-11

7 Conclusions and future directions

In this chapter, a wrapper FS approach (called BDA) that employs SSA as a

selection strategy and KNN as an evaluator was proposed. BDA was tested
using nine large scale medical datasets with low number of samples. Dealing
with high dimensional datasets with low number of samples is challenging
since the model cannot has no enough samples to be trained, in addition
16 Mafarja et. al

0.25
GWO GWO
0.4

Fitness Value

Fitness Value
BGSA BGSA
BBA
0.2 BBA
0.3
BDA BDA
0.2
0.15
0.1

0 0.1
0 25 50 75 100 0 25 50 75 100
Iteration Number Iteration Number
(a) 11 Tumors (b) 14 Tumors
0.6 0.4
GWO GWO
Fitness Value

Fitness Value
BGSA BGSA
0.4 BBA
0.3 BBA
BDA BDA

0.2 0.2

0 0.1
0 25 50 75 100 0 25 50 75 100
Iteration Number Iteration Number
(c) Brain Tumor1 (d) Brain Tumor2

GWO GWO
0.4 0.4
Fitness Value

Fitness Value

BGSA BGSA
0.3 BBA 0.3 BBA
BDA BDA
0.2 0.2

0.1 0.1

0 0
0 25 50 75 100 0 25 50 75 100
Iteration Number Iteration Number
(e) DLBCL (f) Leukemia1
0.2 0.4
GWO GWO
Fitness Value

Fitness Value

0.15 BGSA BGSA

BBA
0.3 BBA
0.1 BDA BDA

0.2
0.05

0 0.1
0 25 50 75 100 0 25 50 75 100
Iteration Number Iteration Number
(g) Leukemia2 (h) Prostate Tumor

GWO
0.14
Fitness Value

BGSA
0.12 BBA
BDA
0.1

0.08

0.06
0 25 50 75 100
Iteration Number
(i) SRBCT

Fig. 5: Convergence curves for BGWO, BGSA, BBA, and BDA for all
datasets.
DA: Theory, Literature Review, and Application in Feature Selection 17

to having a large search space. BDA showed a good performance on those

datasets when compared with five well known wrapper FS approaches.

References

[1] An efficient binary salp swarm algorithm with crossover scheme for fea-
ture selection problems. Knowledge-Based Systems 154, 43 – 67 (2018).
[2] Abdel-Basset, M., Luo, Q., Miao, F., Zhou, Y.: Solving 0-1 knapsack
problems by binary dragonfly algorithm. In: International Conference
on Intelligent Computing. pp. 491–502. Springer (2017)
[3] Al-Madi, N., Aljarah, I., Ludwig, S.: Parallel glowworm swarm opti-
mization clustering algorithm based on mapreduce. In: IEEE Sympo-
sium Series on Computational Intelligence (IEEE SSCI 2014). IEEE
Xplore Digital Library (2014)
[4] Aljarah, I., AlaM, A.Z., Faris, H., Hassonah, M.A., Mirjalili, S., Saadeh,
H.: Simultaneous feature selection and support vector machine opti-
mization using the grasshopper optimization algorithm. Cognitive Com-
putation pp. 1–18 (2018)
[5] Aljarah, I., Faris, H., Mirjalili, S.: Optimizing connection weights in neu-
ral networks using the whale optimization algorithm. Soft Computing
22(1), 1–15 (2018)
[6] Aljarah, I., Faris, H., Mirjalili, S., Al-Madi, N.: Training radial basis
function networks using biogeography-based optimizer. Neural Com-
puting and Applications 29(7), 529–553 (2018)
[7] Aljarah, I., Ludwig, S.A.: Parallel particle swarm optimization cluster-
ing algorithm based on mapreduce methodology. In: In Proceedings of
the Fourth World Congress on Nature and Biologically Inspired Com-
puting (IEEE NaBIC12). IEEE Explore (2012)
[8] Aljarah, I., Ludwig, S.A.: A mapreduce based glowworm swarm op-
timization approach for multimodal functions. In: IEEE Symposium
Series on Computational Intelligence, IEEE SSCI 2013. IEEE Xplore
(2013)
[9] Aljarah, I., Ludwig, S.A.: A new clustering approach based on glowworm
swarm optimization. In: In Proceedings of 2013 IEEE Congress on Evo-
lutionary Computation Conference (IEEE CEC13), Cancun, Mexico.
IEEE Xplore (2013)
[10] Aljarah, I., Ludwig, S.A.: Towards a scalable intrusion detection system
based on parallel pso clustering using mapreduce. In: In Proceedings of
Genetic and Evolutionary Computation Conference (ACM GECCO13)
Amsterdam, July 2013. ACM (2013)
[11] Aljarah, I., Ludwig, S.A.: A scalable mapreduce-enabled glowworm
swarm optimization approach for high dimensional multimodal func-
18 Mafarja et. al

tions. International Journal of Swarm Intelligence Research (IJSIR)

7(1), 32–54 (2016)
[12] Altman, N.S.: An introduction to kernel and nearest-neighbor nonpara-
metric regression. The American Statistician 46(3), 175–185 (1992)
[13] Amroune, M., Bouktir, T., Musirin, I.: Power system voltage stability
assessment using a hybrid approach combining dragonfly optimization
algorithm and support vector regression. Arabian Journal for Science
and Engineering pp. 1–14 (2018)
[14] Babayigit, B.: Synthesis of concentric circular antenna arrays using
dragonfly algorithm. International Journal of Electronics 105(5), 784–
793 (2018)
[15] Barham, R., Aljarah, I.: Link prediction based on whale optimization al-
gorithm. In: The International Conference on new Trends in Computing
Sciences (ICTCS2017), Amman, Jordan. (2017)
[16] Bhesdadiya, R., Pandya, M.H., Trivedi, I.N., Jangir, N., Jangir, P., Ku-
mar, A.: Price penalty factors based approach for combined economic
emission dispatch problem solution using dragonfly algorithm. In: En-
ergy Efficient Technologies for Sustainability (ICEETS), 2016 Interna-
tional Conference on. pp. 436–441. IEEE (2016)
[17] Blum, A.L., Langley, P.: Selection of relevant features and examples in
machine learning. Artificial intelligence 97(1-2), 245–271 (1997)
[18] Branch, S.R., Rey, S.: Providing a load balancing method based on
dragonfly optimization algorithm for resource allocation in cloud com-
puting. International Journal of Networked and Distributed Computing
6(1), 35–42 (2018)
[19] Chen, Y., Li, L., Xiao, J., Yang, Y., Liang, J., Li, T.: Particle swarm op-
timizer with crossover operation. Engineering Applications of Artificial
Intelligence 70, 159–169 (2018)
[20] Daely, P.T., Shin, S.Y.: Range based wireless node localization using
dragonfly algorithm. In: Ubiquitous and Future Networks (ICUFN),
2016 Eighth International Conference on. pp. 1012–1015. IEEE (2016)
[21] Dorigo, M., Birattari, M.: Ant colony optimization. In: Encyclopedia of
machine learning, pp. 36–39. Springer (2011)
[22] Dorigo, M., Di Caro, G.: Ant colony optimization: a new meta-heuristic.
In: Evolutionary Computation, 1999. CEC 99. Proceedings of the 1999
Congress on. vol. 2, pp. 1470–1477. IEEE (1999)
[23] Eberhart, R., Kennedy, J.: A new optimizer using particle swarm theory.
In: Micro Machine and Human Science, 1995. MHS’95., Proceedings of
the Sixth International Symposium on. pp. 39–43. IEEE (1995)
[24] Elhariri, E., El-Bendary, N., Hassanien, A.E.: Bio-inspired optimization
for feature set dimensionality reduction. In: Advances in Computational
Tools for Engineering Applications (ACTEA), 2016 3rd International
Conference on. pp. 184–189. IEEE (2016)
[25] Faris, H., Aljarah, I., Al-Madi, N., Mirjalili, S.: Optimizing the learning
process of feedforward neural networks using lightning search algorithm.
DA: Theory, Literature Review, and Application in Feature Selection 19

International Journal on Artificial Intelligence Tools 25(06), 1650033

(2016)
[26] Faris, H., Aljarah, I., Al-Shboul, B.: A hybrid approach based on particle
swarm optimization and random forests for e-mail spam filtering. In:
International Conference on Computational Collective Intelligence. pp.
498–508. Springer, Cham (2016)
[27] Faris, H., Aljarah, I., Mirjalili, S.: Evolving radial basis function net-
works using moth–flame optimizer. In: Handbook of Neural Computa-
tion, pp. 537–550 (2017)
[28] Faris, H., Aljarah, I., et al.: Optimizing feedforward neural networks
using krill herd algorithm for e-mail spam detection. In: Applied Elec-
trical Engineering and Computing Technologies (AEECT), 2015 IEEE
Jordan Conference on. pp. 1–5. IEEE (2015)
[29] Faris, H., Mafarja, M.M., Heidari, A.A., Aljarah, I., AlaM, A.Z., Mir-
jalili, S., Fujita, H.: An efficient binary salp swarm algorithm with
crossover scheme for feature selection problems. Knowledge-Based Sys-
tems 154, 43–67 (2018)
[30] Fisher, L.: The perfect swarm: The science of complexity in everyday
life. Basic Books (2009)
[31] Ghatasheh, N., Faris, H., Aljarah, I., Al-Sayyed, R.M.: Optimizing soft-
ware effort estimation models using firefly algorithm. Journal of Soft-
ware Engineering and Applications 8(03), 133 (2015)
[32] Guha, K., Laskar, N., Gogoi, H., Borah, A., Baishnab, K., Baishya, S.:
Novel analytical model for optimizing the pull-in voltage in a flexured
mems switch incorporating beam perforation effect. Solid-State Elec-
tronics 137, 85–94 (2017)
[33] Guo, S., Dooner, M., Wang, J., Xu, H., Lu, G.: Adaptive engine op-
timisation using nsga-ii and moda based on a sub-structured artificial
neural network. In: Automation and Computing (ICAC), 2017 23rd In-
ternational Conference on. pp. 1–6. IEEE (2017)
[34] Hamdy, M., Nguyen, A.T., Hensen, J.L.: A performance comparison of
multi-objective optimization algorithms for solving nearly-zero-energy-
building design problems. Energy and Buildings 121, 57–71 (2016)
[35] Hariharan, M., Sindhu, R., Vijean, V., Yazid, H., Nadarajaw, T., Yaa-
cob, S., Polat, K.: Improved binary dragonfly optimization algorithm
and wavelet packet based non-linear features for infant cry classification.
Computer Methods and Programs in Biomedicine 155, 39–51 (2018)
[36] Heidari, A.A., Kazemizade, O., Hakimpour, F.: A new hybrid yin-yang-
pair swarm optimization algorithm for uncapacitated warehouse loca-
tion problems. ISPRS - International Archives of the Photogrammetry,
Remote Sensing and Spatial Information Sciences XLII-4/W4, 373–
379 (2017).
[37] Heidari, A.A., Mirvahabi, S.S., Homayouni, S.: An effective hybrid sup-
port vector regression with chaos-embedded biogeography-based opti-
mization strategy for prediction of earthquake-triggered slope deforma-
20 Mafarja et. al

tions. ISPRS - International Archives of the Photogrammetry, Remote

Sensing and Spatial Information Sciences XL-1/W5, 301–305 (2015).
[38] Heidari, A.A., Moayedi, A., Ali Abbaspour, R.: Estimating origin-
destination matrices using an efficient moth flame-based spatial cluster-
ing approach. ISPRS - International Archives of the Photogrammetry,
Remote Sensing and Spatial Information Sciences XLII-4/W4, 381–
387 (2017).
[39] Heidari, A.A., Abbaspour, R.A.: Enhanced chaotic grey wolf optimizer
for real-world optimization problems: A comparative study. In: Hand-
book of Research on Emergent Applications of Optimization Algo-
rithms, pp. 693–727. IGI Global (2018)
[40] Heidari, A.A., Abbaspour, R.A., Jordehi, A.R.: An efficient chaotic wa-
ter cycle algorithm for optimization tasks. Neural Computing and Ap-
plications 28(1), 57–85 (2017)
[41] Heidari, A.A., Abbaspour, R.A., Jordehi, A.R.: Gaussian bare-bones
water cycle algorithm for optimal reactive power dispatch in electrical
power systems. Applied Soft Computing 57, 657–671 (2017)
[42] Heidari, A.A., Delavar, M.R.: A modified genetic algorithm for find-
ing fuzzy shortest paths in uncertain networks. ISPRS - International
Archives of the Photogrammetry, Remote Sensing and Spatial Informa-
tion Sciences XLI-B2, 299–304 (2016).
[43] Heidari, A.A., Pahlavani, P.: An efficient modified grey wolf optimizer
with lévy flight for optimization tasks. Applied Soft Computing 60,
115–134 (2017)
[44] Hema, C., Sankar, S., et al.: Energy efficient cluster based protocol to
extend the rfid network lifetime using dragonfly algorithm. In: Commu-
nication and Signal Processing (ICCSP), 2016 International Conference
on. pp. 0530–0534. IEEE (2016)
[45] Hema, C., Sankar, S., et al.: Performance comparison of dragonfly and
firefly algorithm in the rfid network to improve the data transmission.
Journal of Theoretical and Applied Information Technology 95(1), 59
(2017)
[46] Hema Sekhar, A., Lakshmi Devi, A.: Analysis of multi tcsc placement in
transmission system by using firing angle control model with heuristic
algorithms. ARPN Journal of Engineering and Applied Sciences 11(21),
12743–12755 (2016), cited By 0
[47] Ibrahim, H.T., Mazher, W.J., Ucan, O.N., Bayat, O.: A grasshopper
optimizer approach for feature selection and optimizing svm parameters
utilizing real biomedical data sets. Neural Computing and Applications
(Mar 2018).
[48] Jafari, M., Chaleshtari, M.H.B.: Using dragonfly algorithm for opti-
mization of orthotropic infinite plates with a quasi-triangular cut-out.
European Journal of Mechanics-A/Solids 66, 1–14 (2017)
DA: Theory, Literature Review, and Application in Feature Selection 21

[49] Karaboga, D., Basturk, B.: A powerful and efficient algorithm for nu-
merical function optimization: artificial bee colony (abc) algorithm.
Journal of global optimization 39(3), 459–471 (2007)
[50] Kashef, S., Nezamabadi-pour, H.: An advanced aco algorithm for feature
subset selection 147, 271–279 (2015)
[51] Kennedy, J.: Swarm intelligence. In: Handbook of nature-inspired and
innovative computing, pp. 187–219. Springer (2006)
[52] Khadanga, R.K., Padhy, S., Panda, S., Kumar, A.: Design and analysis
of tilt integral derivative controller for frequency control in an islanded
microgrid: A novel hybrid dragonfly and pattern search algorithm ap-
proach. Arabian Journal for Science and Engineering pp. 1–12 (2018)
[53] KS, S.R., Murugan, S.: Memory based hybrid dragonfly algorithm for
numerical optimization problems. Expert Systems with Applications 83,
63–78 (2017)
[54] Kumar, C.A., Vimala, R., Britto, K.A., Devi, S.S.: Fdla: Fractional
dragonfly based load balancing algorithm in cluster cloud model. Cluster
Computing pp. 1–14 (2018)
[55] Langley, P., et al.: Selection of relevant features in machine learning.
In: Proceedings of the AAAI Fall symposium on relevance. vol. 184, pp.
245–271 (1994)
[56] Li, Q., Chen, H., Huang, H., Zhao, X., Cai, Z., Tong, C., Liu, W., Tian,
X.: An enhanced grey wolf optimization based feature selection wrapped
kernel extreme learning machine for medical diagnosis. Computational
and mathematical methods in medicine 2017 (2017)
[57] Li, Y., Li, T., Liu, H.: Recent advances in feature selection and its ap-
plications. Knowledge and Information Systems 53(3), 551–577 (2017)
[58] Liao, T., Kuo, R.: Five discrete symbiotic organisms search algorithms
for simultaneous optimization of feature subset and neighborhood size
of knn classification models. Applied Soft Computing 64, 581 – 595
(2018).
[59] Liu, H., Motoda, H.: Feature selection for knowledge discovery and data
mining, vol. 454. Springer Science & Business Media (2012)
[60] Liu, H., Setiono, R., et al.: A probabilistic approach to feature selection-
a filter solution. In: Thirteenth International Conference on Machine
Learning (ICML). vol. 96, pp. 319–327. Citeseer (1996)
[61] Mafarja, M., Abdullah, S.: Modified great deluge for attribute reduc-
tion in rough set theory. In: Fuzzy Systems and Knowledge Discovery
(FSKD), 2011 Eighth International Conference on. vol. 3, pp. 1464–
1469. IEEE (2011)
[62] Mafarja, M., Abdullah, S.: Investigating memetic algorithm in solving
rough set attribute reduction. International Journal of Computer Ap-
plications in Technology 48(3), 195–202 (2013)
[63] Mafarja, M., Abdullah, S.: Record-to-record travel algorithm for at-
tribute reduction in rough set theory. J Theor Appl Inf Technol 49(2),
507–513 (2013)
22 Mafarja et. al

[64] Mafarja, M., Abdullah, S.: Fuzzy modified great deluge algorithm for
attribute reduction. In: Recent Advances on Soft Computing and Data
Mining, pp. 195–203. Springer, Cham (2014)
[65] Mafarja, M., Abdullah, S.: A fuzzy record-to-record travel algorithm for
solving rough set attribute reduction. International Journal of Systems
Science 46(3), 503–512 (2015)
[66] Mafarja, M., Aljarah, I., Heidari, A.A., Hammouri, A.I., Faris, H.,
AlaM, A.Z., Mirjalili, S.: Evolutionary population dynamics and
grasshopper optimization approaches for feature selection problems.
Knowledge-Based Systems 145, 25 – 45 (2018).
[67] Mafarja, M., Eleyan, D., Abdullah, S., Mirjalili, S.: S-shaped vs. v-
shaped transfer functions for ant lion optimization algorithm in feature
selection problem. In: Proceedings of the International Conference on
Future Networks and Distributed Systems. p. 14. ACM (2017)
[68] Mafarja, M., Jaber, I., Eleyan, D., Hammouri, A., Mirjalili, S.: Binary
dragonfly algorithm for feature selection. In: 2017 International Confer-
ence on New Trends in Computing Sciences (ICTCS). pp. 12–17 (2017).
[69] Mafarja, M., Mirjalili, S.: Hybrid whale optimization algorithm with
simulated annealing for feature selection. Neurocomputing (2017)
[70] Mafarja, M., Mirjalili, S.: Whale optimization approaches for wrapper
feature selection. Applied Soft Computing 62, 441–453 (2017)
[71] Mirjalili, S.: The ant lion optimizer. Advances in Engineering Software
83, 80–98 (2015)
[72] Mirjalili, S.: Dragonfly algorithm: a new meta-heuristic optimization
technique for solving single-objective, discrete, and multi-objective
problems. Neural Computing and Applications 27(4), 1053–1073 (2016).
[73] Mirjalili, S., Gandomi, A.H., Mirjalili, S.Z., Saremi, S., Faris, H., Mir-
jalili, S.M.: Salp swarm algorithm: a bio-inspired optimizer for engineer-
ing design problems. Advances in Engineering Software 114, 163–191
(2017)
[74] Mirjalili, S., Lewis, A.: S-shaped versus v-shaped transfer functions for
binary particle swarm optimization. Swarm and Evolutionary Compu-
tation 9, 1–14 (2013).
[75] Mirjalili, S., Lewis, A.: The whale optimization algorithm. Advances in
Engineering Software 95, 51–67 (2016)
[76] Mirjalili, S., Mirjalili, S.M., Lewis, A.: Grey wolf optimizer. Advances
in engineering software 69, 46–61 (2014)
[77] Moradi, P., Gholampour, M.: A hybrid particle swarm optimization for
feature subset selection by integrating a novel local search strategy.
Applied Soft Computing 43, 117–130 (2016)
[78] Pathania, A.K., Mehta, S., Rza, C.: Economic load dispatch of wind
thermal integrated system using dragonfly algorithm. In: Power Elec-
tronics (IICPE), 2016 7th India International Conference on. pp. 1–6.
IEEE (2016)
DA: Theory, Literature Review, and Application in Feature Selection 23

[79] Rajamohana, S., Umamaheswari, K.: Hybrid approach of improved bi-

nary particle swarm optimization and shuffled frog leaping for feature
selection. Computers & Electrical Engineering (2018)
[80] Raman, G., Raman, G., Manickam, C., Ganesan, S.I.: Dragonfly al-
gorithm based global maximum power point tracker for photovoltaic
systems. In: Tan, Y., Shi, Y., Niu, B. (eds.) Advances in Swarm Intel-
ligence. pp. 211–219. Springer International Publishing, Cham (2016)
[81] Reynolds, C.W.: Flocks, herds and schools: A distributed behavioral
model. In: ACM SIGGRAPH computer graphics. vol. 21, pp. 25–34.
ACM (1987)
[82] Salam, M.A., Zawbaa, H.M., Emary, E., Ghany, K.K.A., Parv, B.: A
hybrid dragonfly algorithm with extreme learning machine for predic-
tion. In: INnovations in Intelligent SysTems and Applications (INISTA),
2016 International Symposium on. pp. 1–6. IEEE (2016)
[83] Sambandam, R.K., Jayaraman, S.: Self-adaptive dragonfly based opti-
mal thresholding for multilevel segmentation of digital images. Journal
of King Saud University-Computer and Information Sciences (2016)
[84] Saremi, S., Mirjalili, S., Lewis, A.: Grasshopper optimisation algorithm:
Theory and application. Advances in Engineering Software 105, 30–47
(2017)
[85] Shilaja, C., Ravi, K.: Optimal power flow using hybrid da-apso algo-
rithm in renewable energy resources. Energy Procedia 117, 1085–1092
(2017)
[86] Shukri, S., Faris, H., Aljarah, I., Mirjalili, S., Abraham, A.: Evolution-
ary static and dynamic clustering algorithms based on multi-verse op-
timizer. Engineering Applications of Artificial Intelligence 72, 54–66
(2018)
[87] Song, J., Li, S.: Elite opposition learning and exponential function steps-
based dragonfly algorithm for global optimization. In: Information and
Automation (ICIA), 2017 IEEE International Conference on. pp. 1178–
1183. IEEE (2017)
[88] Statnikov, A., Aliferis, C.F., Tsamardinos, I., Hardin, D., Levy, S.: A
comprehensive evaluation of multicategory classification methods for
microarray gene expression cancer diagnosis. Bioinformatics 21(5), 631–
643 (2004)
[89] Sugave, S.R., Patil, S.H., Reddy, B.E.: Ddf: Diversity dragonfly algo-
rithm for cost-aware test suite minimization approach for software test-
ing. In: Intelligent Computing and Control Systems (ICICCS), 2017
International Conference on. pp. 701–707. IEEE (2017)
[90] Suresh, V., Sreejith, S.: Generation dispatch of combined solar thermal
systems using dragonfly algorithm. Computing 99(1), 59–80 (2017)
[91] Surowiecki, J., Silverman, M.P., et al.: The wisdom of crowds. American
Journal of Physics 75(2), 190–192 (2007)
[92] Tharwat, A., Gabel, T., Hassanien, A.E.: Classification of toxicity ef-
fects of biotransformed hepatic drugs using optimized support vector
 24 Mafarja et. al

machine. In: Hassanien, A.E., Shaalan, K., Gaber, T., Tolba, M.F.
(eds.) Proceedings of the International Conference on Advanced Intelli-
gent Systems and Informatics 2017. pp. 161–170. Springer International
Publishing, Cham (2017)
[93] Tharwat, A., Gabel, T., Hassanien, A.E.: Parameter optimization of
support vector machine using dragonfly algorithm. In: International
Conference on Advanced Intelligent Systems and Informatics. pp. 309–
319. Springer (2017)
[94] Vanishree, J., Ramesh, V.: Optimization of size and cost of static var
compensator using dragonfly algorithm for voltage profile improvement
in power transmission systems. International Journal of Renewable En-
ergy Research (IJRER) 8(1), 56–66 (2018)
[95] VeeraManickam, M., Mohanapriya, M., Pandey, B.K., Akhade, S., Kale,
S., Patil, R., Vigneshwar, M.: Map-reduce framework based cluster ar-
chitecture for academic students performance prediction using cumula-
tive dragonfly based neural network. Cluster Computing pp. 1–17 (2018)
[96] Vikram, K.A., Ratnam, C., Lakshmi, V., Kumar, A.S., Ramakanth, R.:
Application of dragonfly algorithm for optimal performance analysis of
process parameters in turn-mill operations-a case study. In: IOP Con-
ference Series: Materials Science and Engineering. vol. 310, p. 012154.
IOP Publishing (2018)
[97] Wu, J., Zhu, Y., Wang, Z., Song, Z., Liu, X., Wang, W., Zhang, Z., Yu,
Y., Xu, Z., Zhang, T., et al.: A novel ship classification approach for
high resolution sar images based on the bda-kelm classification model.
International Journal of Remote Sensing 38(23), 6457–6476 (2017)
[98] Zorarpacı, E., Özel, S.A.: A hybrid approach of differential evolution
and artificial bee colony for feature selection. Expert Systems with Ap-
plications 62, 91–103 (2016)

View publication stats