Ingénierie des Systèmes d’Information

Vol. 24, No. 3, June, 2019, pp. 281-287

Journal homepage: http://iieta.org/journals/isi

A Hybrid Modified Whale Optimization Algorithm with Simulated Annealing for Terrorism
Prediction
Ghada M.A. Soliman1*, Tarek H.M. Abou-El-Enien1, Eid Emary2, Motaz M.H. Khorshid1
1
Faculty of Computers & Artificial Intelligence, Cairo University, Giza, Egypt
2
Faculty of Computer Studies, Arab Open University, Cairo, Egypt

Corresponding Author Email: gh.tolan@ fci-cu.edu.eg

https://doi.org/10.18280/isi.240308 ABSTRACT

Received: 1 April 2019 The main purpose of this research is to propose a novel hybrid memetic searching optimization
Accepted: 7 June 2019 algorithm, the proposed algorithm integrates the modified whale optimization algorithms with
simulated annealing algorithm in a novel approach in order to enhance the searching
Keywords: capabilities of the modified whale optimization algorithm with the help of simulated annealing
hybrid algorithms, memetic algorithm, (SA). The proposed algorithm is used to find the minimum feature subset based on hybrid
whale optimization algorithm, feature modified whale optimization algorithms and simulated annealing (WOA2SA, WOA3SA),
selection, spiral path, tournament where SA is embedded into the modified WO algorithms to achieve that good balance between
selection (exploitation) and (exploration) capabilities of the modified algorithms. The resulting memetic
algorithm improve the performance of the general classification tasks and hence had been used
in the prediction of terrorist group (s) which responsible of the terror attacks on Egypt based
on GTD terrorism data. The findings of this research can serve as an alarm tool to minimize
the terrorist attacks on a certain region (country).

1. INTRODUCTION because the major strength of population-based methods in

their exploration capability and the strength of local search
In combinatorial optimization (NP-hard) problems; methods in their rather fast intensification capability; so the
computing optimal solutions is computationally intractable, hybrid algorithms can provide more efficient behavior and
and so most researchers are usually satisfied with what known higher flexibility, in the field of evolutionary algorithms these
as “good” solutions that are obtained by metaheuristic hybrid algorithms are known as own name, memetic
algorithms. In addition to single solution search algorithms algorithms [10].
such as simulated annealing (SA) [1], tabu search [2], greedy In this research; a novel hybrid memetic algorithm are
heuristic [3], there is a growth interest in population based presented that based on hybridization between the modified
metaheuristics especially evolutionary algorithms (EA) such whale optimization algorithms (WOA2, WOA3) with the
as genetic algorithms (GA) [4], genetic programming [5], ant simulated annealing into a classification mechanism to provide
colonies (AC) [6], scatter search (SS) [7], moth-flame a prediction system for one of the NP-hard real problems with
optimization algorithm [8] and other numerous of new high performance. The hybrid memetic algorithm proved its
algorithms. capability to enhance the classification performance and
The hybridization of metaheuristics rises as a new research feature selection with highly accuracy and hence used in the
field in optimization due to the fact that the focus of research prediction of the terrorist group(s) whose responsible of the
has shifted from an algorithm-oriented point of view to a terror attacks on Egypt.
problem-oriented point of view. The research in the last years The remainder of this paper is organized as follows: Section
have produced a large number of algorithms that simply do not 2 presents an overview about the simulated annealing
fit into a single metaheuristic category. This is because these algorithm, while Section 3 explains in details the modified
untraditional research approaches combine various whale optimization algorithms. In Section 4; the proposed
algorithmic ideas, mostly originated from several branches of hybrid memetic algorithm components are introduced as well
artificial intelligence, operations research and computer as the implementation, the results and analysis of the model
science. Such approaches are referred to as hybrid are presented. Section 5 gives the main conclusion of the
metaheuristics [9]. The hybrid algorithms have the ability for research and further points for future research.
improving run time, results or both. Research on hybrid
metaheuristics can be subdivided into five different categories,
namely, the hybridization of metaheuristics with (meta-) 2. SIMULATED ANNEALING
heuristics which will be our main focus in this research. The
other categories are constraint programming, tree search Simulated Annealing (SA) is a random and local search
methods, problem relaxations, and dynamic programming. technique [11] was developed in order to solve the difficult
The hybridization of metaheuristics with (meta-) heuristics and highly nonlinear optimization problems. SA is a robust
is quite popular, especially whose concentrates on using of and general techniques as it has the ability to deal with many
local search methods inside population-based methods; constraints' problems, the noisy and chaotic data problems. SA

281
based on using the highly similarity between the cooling and prey, as well as using the spiral helix shape movement in
freezing way in which a metal cools and freezes into a hunting. The spiral shaped path that explained in the original
minimum energy crystalline structure that is called as "The WO algorithm is follows a logarithmic shape also called
Annealing", and the search for a minimum solution in a more (equiangular spiral) ) but the constant property of this spiral
general system; SA is a basis that allow an optimization function is not realistic for the whales while its movement (the
technique to solve the combinatorial problems as it can deal constant property in which the angle formed by the radial
with other different disciplines problems. SA has a very good vector, this vector is formed from a line between any point
ability to escape from the local optima in which the improving "W" on the spiral toward the center of the spiral is constant )
move will be accepted but the worse one will be accepted as a which is not accurate to represent the whales movement in
solution under determined probability. The probability of reality, and hence we intended in our research in choosing a
accepting the worse solution is calculated by using the more accurate and realistic spiral shape after investigate
Boltzmann probability as follows in Eq. (1). different forms of spiral shaped to simulate the helix-
movement of the whales that enhance the exploitation
−∝⁄
Probability = e T emper (1) capability of humpback whales in the hunting mechanism.

where ∝ : objective difference evaluation between best

solution (𝑆𝑜𝑙𝑛𝑏𝑒𝑠𝑡 ) and the trail solution (𝑆𝑜𝑙𝑛𝑡𝑟𝑖𝑎𝑙 ), Temper:
Temperature that periodically decreases during the search
process.
And hence we can conclude that SA has main advantages
over other local search algorithms in its high flexibility and the
capability to approach global optimality as well as it does not
depend on a restrictive structure that makes the algorithm in
Figure 1 is quite versatile and can easily tuned to be used with
other searching algorithms.

Figure 1. Simulated annealing algorithm

Figure 2. Bubble-Net haunting behavior of humpback
whales
3. MODIFIED WHALE OPTIMIZATION
ALGORITHMS (WOA2, WOA3) The new spiral shaped path that have been used in our
research known as hyperbolic (reciprocal) and Archimedean
In Whale optimization algorithm [12]; the humpback
spiral which proposed by Soliman et al. in [13] then investigate
whales which are special kind of whales hunt a school of krill
the performance of the modified versions of WO algorithm by
or small fishes in which they are in a near area to the surface
conducting different experiments within a hybrid terrorism
by creating a trap consists from bubbles called bubble-net
prediction system to predict the terrorist group(s) responsible
strategy. The humpback whales have the ability to swim down
of attacks on Egypt along the time period (from 1996 to 2017).
in water in approximate of (10-15 meters) around those fishes
The results of the modified WOA is tested, evaluated, then
with creating distinctive bubbles following a helix-shaped
compared with the original algorithm, other two modified
movement. The helix-9 shaped is defined as a spiral shape
versions of the modified Moth flame optimization algorithm
along a circle. Then the prey can follow the bubbles and moves
(MFO) and its modified versions (MFO2, MFO3) that have
upward the surface to be hunted by whales, as showed in
been proposed by Soliman et al in [14], and the popular
Figure 2. The whales have different abilities allow it to be one
algorithms such as Particle Swarm, and Genetic Algorithms.
of the promising algorithms and more powerful, efficient
The main phases of WOA2, and WOA3; exploration and
algorithm to lead to the best global search such as the ability
exploitation phases are presented below:
to operate on a random basis or best agent while chasing the

282
3.1 Exploitation phase Archimedes’ Spiral:
⃗⃗⃗⃗⃗⃗⃗⃗ ⃗ . r cos( 2πr) + ⃗⃗⃗⃗⃗⃗⃗
WP (c + 1) = ⃗D WP ∗ (𝑐) (8)
3.1.1 Shrinking encircling a prey
The humpback whales begin to encircle the prey near the
surface of the water as described mathematically by Eq. (2) where
and Eq. (3) and explained in [9, 12, 13].
⃗́ = |W
⃗⃗⃗⃗⃗⃗⃗∗ ⃗⃗⃗⃗⃗
𝐷 P (c) − WP (𝑐) |: is the distance do the whale number
⃗⃗ p (c + 1) = ⃗⃗⃗⃗⃗⃗⃗
⃗W Wp ∗ (c) − ⃗M
⃗⃗ . ⃗D
⃗ (2) j- th whale to the prey (the last obtained best solution), and r
random number in [−1, 1].
D ⃗⃗ . ⃗⃗⃗⃗⃗⃗⃗
⃗⃗ = | N WP ∗ (n) − ⃗⃗⃗⃗⃗
Wp (c)| (3)
The representation of the two hunting mechanisms of the
whales (shrinking encircling and the bubble-net by spiral-
⃗⃗⃗⃗
M = 2 ⃗⃗⃗⃗
M . s − ⃗⃗⃗⃗
M (4) shape path), is by considering a 50 % probability to make the
choice between them in the optimization process as explained
⃗⃗⃗N = 2 s (5) in the Eq. (9).

where: ⃗⃗⃗⃗⃗
WP (c + 1)
⃗⃗⃗⃗⃗⃗
𝑊 ∗ (c) − M. ⃗⃗⃗⃗ D
⃗⃗ if (A < 0.5)
⃗⃗⃗⃗⃗
Wp : The whale's position vector. cos( 2πr)
= ⃗⃗́ .
D + ⃗⃗⃗⃗⃗
W ∗ (c) if (A ≥ 0.5)
⃗⃗⃗⃗⃗⃗⃗
WP ∗ : Represents historically best whales' position (solution) r
obtained so far. ⃗ . r cos( 2πr) + ⃗⃗⃗⃗⃗⃗⃗
⃗D WP ∗ (c)
{
c: indicates current iteration. (9)
⃗⃗⃗⃗
M , ⃗N ⃗ : known as coefficient vectors that are calculated as
in Eq. (4) and Eq. (5) respectively. where: A considered a random number that defined in [0, 1].
⃗⃗⃗⃗ : Called distance control parameter, its values is decreased
M
from 2 till 0 within the iteration number. 3.3 Exploration (diversification)
s ∶ A random vector follows the uniform distribution in the
interval of [0, 1]. The humpback whales can search for the prey via exploring
As explained by Eq. (2) the whales enhance their positions the area globally and to enhance this exploration ability; they
according to the position of the best known solution (optimum based on searching for the prey by a random selected whale to
solution). Whales can be located in the neighborhood of small guide the search which is not depends on the whales' positions
fishes via changing the values of ⃗⃗⃗⃗ M and ⃗N ⃗ vectors; The to each other. This mechanism is completely differs from the
Shrinking encircling behavior of a whale around the prey is exploitation phase in which the whales can enhance and update
achieved according to the decreasing value of M ⃗⃗⃗⃗ in Eq. (4) that their position depending on the best whales' position obtained
explained by Eq. (6) so far, and hence, a vector M can take values "greater than 1"
or "less than −1" is used to allow the whales to move far away
⃗⃗⃗⃗ = 2 − c 2 from the best whale position. The mathematical model for the
M (6)
MaxIn exploration mechanism is as follows in Eq. (10) amd Eq. (11):

where ⃗D
⃗ = |N
⃗⃗⃗⃗ . ⃗⃗⃗⃗⃗⃗
WP random − ⃗⃗⃗⃗⃗⃗
WP | (10)
c: the number of the iteration.
MaxIn: maximum iterations.
⃗⃗⃗⃗
WP (c + 1) = ⃗⃗⃗⃗⃗ ⃗⃗⃗ . D
WP random − M ⃗⃗ (11)
3.1.2 Bubble-net attacking method
To determine the position of neighbor around the whale; the where
distance between a whale position (W ⃗⃗⃗⃗⃗⃗P ) and the best known ⃗⃗⃗⃗⃗⃗
WP random : a random position vector represents a random
⃗⃗⃗⃗⃗⃗⃗∗
whale position obtained so far (WP ) has to be calculated. This chosen humpback whale.
distance is represented by spiral path called hyperbolic The Following steps present the modified whale
(reciprocal) spiral that we based on constructing the spiral path optimization algorithms (WOA2, WOA3). The algorithm
for our modified WO algorithm (WOA2) and Archimedes’ begins with creating and initializing a set of humpback whales
Spiral that we based on in constructing the spiral function for by using the uniform initialization method, then once the
(WOA3). These two spiral functions are illustrated and optimization starts; the position evaluation of these whales are
defined in [15]. done based on the defined fitness. The algorithm continue
proceeding till finds the best whale position around the small
3.2 Mathematical formulation fishes (the prey), the algorithm repeatedly executes the
following steps until the maximum iteration reached.
The Spiral paths equations that represent the humpback
whales movement for WOA2 as explained by Eq. (7) and for Step 1: Update the main algorithm coefficients.
WOA3 defined below in Eq. (8) as follows: Step 2: a random value is generated that used to determine
Hyperbolic Spiral: how the whales updates their position by using either one of
the two mechanisms represented by Eq. (2) or Eq. (11) or
based on the spiral paths explained in Eq. (7) and Eq. (8).
WP (c + 1) = ⃗⃗D́.
cos( 2πr)
⃗⃗⃗⃗⃗⃗⃗⃗ + ⃗⃗⃗⃗⃗⃗⃗
WP ∗ (c) (7)
r

283
Step 3: The solutions (whales' positions) cannot go outside the prediction of terrorist group (s) responsible of the terror attacks
search space. on Egypt. In the resulting memetic hybrid algorithm the SA
Step 4: The algorithm presents the output as the best whale's algorithm is selected due to its high advantage in selection that
position with respect to the prey as an approximation to the embedded into (WOA2) and (WOA3) due to their ability in
global solution. global search to achieve that good balance between
(exploitation) and (exploration) abilities of the two algorithms.
The hybridization between the modified (WOA2 with SA)
4. PROPSED HYBRID MODIFIED WHALE and (WOA3 with SA) based on using the "Tournament
OPTIMIZATION SIMULATED ANNEALING selection methodology" of SA algorithm instead of random
ALGORITHMS (WOA2SA, WOA3SA) selection mechanism in the exploration phase of Modified
algorithms (WOA2) and (WOA3) as showed in Figure 3. The
This section presents the proposed hybrid metaheuristics original WOA as well as the two modified versions (WOA2)
algorithm between the Modified Whale optimization and (WOA3) are using a blind operator to play the role of
algorithms (WOA2, WOA3) with the Simulated Annealing exploration or searching for the small fishes (prey), and hence
(SA). The proposed hybrid model is used to find the minimum we used the tournament selection operator in place of the
feature subset that used then to improve the performance of random search which will be considered a solution as its initial
general classification tasks, and hence can perform the state, and replace the original solution by the enhanced one.

Figure 3. The hybrid memetic (WOA2SA) or (WOA3SA) algorithm

284
 Then the Eq. (10) and Eq. (11) will be replaced with the classification process and it passed on multiple steps as
enhanced Eq. (12) and Eq. (13) as follows: explained below:
(1) Data Cleaning: Pre-process data in order to reduce noise
⃗D
⃗ = |N
⃗⃗ ⃗⃗⃗⃗⃗⃗
WP tournament−selection − ⃗⃗⃗⃗⃗
WP | (12) and handle missing values by applying (Litwise-Deletion,
Mode-Imputation) approaches.
⃗⃗⃗⃗⃗ (2) Data transformation (from Categorical into Numeric
WP (c + 1) = ⃗⃗⃗⃗⃗⃗ ⃗⃗⃗ . D
WP tournament−selection − M ⃗⃗ (13)
based on GTD coding system)
(3) The features in our data are classified as (Date, Incident
where location, Location Details, Attack Iinformation, Weapon
⃗⃗⃗⃗⃗⃗
WP 𝒕𝒐𝒖𝒓𝒏𝒂𝒎𝒆𝒏𝒕−𝒔𝒆𝒍𝒆𝒄𝒕𝒊𝒐𝒏 : is a position vector (represents the Information, Target / Victim Information, Casualties &
whale's position) chosen from the tournament selection Consequences) features
process of the SA Algorithm. (4) For the time domain features; we applied a feature
reduction where we transformed the "day, month, and yea"
4.1 Data pre-processing features into an equivalent "day per year, Hijiri day per year,
and an equivalent day per week".
This section explains how the terrorism data has been used (5) For the position or location feature we transformed it
and pre-processed in the proposed prediction model. The used into city-latitude, and city-longitude.
data is real terrorism data and describes the terrorist attacks on (6) Due to the huge number of classes in some features in
Middle East & North Africa especially the attacks that our data; we had to combine the classes in some features into
happened in Egypt (from year 1996 till year 2017) via the groups.
defined fitness function ACC which evaluate the accuracy of The terrorism data on the time period from year 1996 till
the model and explained in Eq. (14). year 2017 has 45 predictors (attributes or features) besides the
The data set derived from the Global Terrorism Database feature "Terrorist group name" which is considered as
(GTD) which is taken from an open source of the National (response class) that we should predict in our research; after
Consortium for the Study of Terrorism and Responses to data-preprocessing steps; the total number of attacks (records)
Terrorism (START) [16]: became 813, the total set of features became 46 features as
The data are required to be prepared for using in the described below in Table 1.

Table 1. Terrorism dataset features for the time period (1996 till 2017) of Egypt

Feature Type Feature Type

Month Numeric Vicinity Categorical
Day Numeric Crit1 (Inclusion Criteria1) Categorical
Year Numeric Crit 2(Inclusion Criteria2) Categorical
Equivelant-Day Per Year Numeric Crit 3(Inclusion Criteria3) Categorical
Equivalent Hijiri day/year Numeric Doubtterr Categorical
Equivalent week per year Numeric Alternative Categorical
Equivalent day per week Numeric Multiple Categorical
Extended Incident Categorical Success) Categorical
Provstate Text Suicide Categorical
City Text Attack-Type Categorical
City-Latitude Numeric TargetType Categorical
City-Longitude Numeric Target-subtype Categorical
Specific Categorical Weapon-Type Categorical
Weapon-Subtype Categorical Compclaim Categorical
Nkill Numeric Property Categorical
Nkilter Numeric Propextent Categorical
Nwound
Numeric Ishostkid Categorical
Nwoundte Numeric Nhostkid Numeric
Guncertain Categorical Nhours Numeric
Individual Categorical Ndays Numeric
Nperps Numeric Rnsome Categorical
nperpcap Numeric Ransomamt Numeric
Claimed Categorical Ransompaid Numeric
Claimmode Categorical Hostkidoutcom Categorical
Victim-Nationality Categorical Group-Name Text

4.2 Fitness function is designed to combine the accuracy of classifier with the
length of feature subset. A fitness function combines both the
Fitness function is important for metaheuristic algorithms, classification error (Classification accuracy) and the number
it is used evaluate the quality of each subset. fitness evaluation of selected features and hence achieve that balance between

285
the classification performance and the reduction size . Table 3. The results of the memetic algorithms with 7 whales

Maximize ACC _Fitness = CCR(C) Number

of Exe.
𝑵𝑪 Mean Selected Time
CCR (C) = (14) Iterations Alg. Fitness ACC. features (Sec.)
𝑵
where 5o Iter. WOA 0.309281 0.692 23 4
WOASA 0.287419 0.613 17 209
CCR(C): is the classification performance measure that
WOA2SA 0.290318 0.619 19 300
represents the correct classification ratio.
NC : is the number of correctly classified data instances and WOA3SA 0.294525 0.640 16 297
N: is the total number of instances in the data set. 70 Iter. WOA 0.323687 0.679 26 6
WOASA 0.271408 0.611 14 257
4.3 Results and analysis
WOA2SA 0.292869 0.518 15 229
The implementation of the hybrid memetic algorithms WOA3SA 0.288599 0.560 16 228
WOASA, WOA2SA, WOA3SA are done using MATLAB 100 Iter. WOA 0.277410 0.723 13 7
R2015a, based on a set of assessment statistical indicators via
WOASA 0.275210 0.587 20 314
different experiments as follows:
Table 2 presents the accuracy and mean fitness as well as WOA2SA 0.273454 0.586 23 321
the execution time for the memetic hybrid algorithms in WOA3SA 0.277414 0.664 16 328
different experiments with 5 searching agents (whales), in 150 Iter. WOA 0.265170 0.72 21 11
which we can observe that as increasing the number of
iterations; the hybrid WOA2SA and WOA3SA perform better WOASA 0.270952 0.684 19 493
than the original WOA and better than the memetic (WOASA) WOA2SA 0.263168 0.556 23 474
in the accuracy of the algorithm as well as in the number of WOA3SA 0.280909 0.5286 15 468
selected features.
Table 3 and Table 4 illustrate how the results are enhanced
by increasing the number of search agents of the hybrid 5. CONCLUSION AND FUTURE RESEARCH
memetic algorithms where the mean fitness and the accuracy
of the modified (WOA2), and (WOA3) with SA are better than In our research paper we proposed a novel hybrid memetic
the original WOA as well as the results of the hybrid searching optimization algorithm, the proposed algorithm
algorithms WOA2SA and WOA3SA are better than WOASA integrates the modified whale optimization algorithms WOA2,
and giving promising performance than WOA. and WOA3 with simulated annealing algorithm in a novel
approach in order to enhance the searching capabilities of the
Table 2. The Results of the hybrid memetic algorithms with modified whale optimization algorithm with the help of
5 whales simulated annealing (SA).
The proposed hybrid memetic algorithm is used to find the
Number minimum feature subset with highest accuracy and maximum
of Exe. fitness based on hybrid modified whale optimization
Mean Selected Time
algorithms and simulated annealing (WOA2SA, WOA3SA),
Iteration Alg. Fitness ACC. features (Sec.)
where SA is embedded into the modified whale optimization
5o Iter. WOA 0.272975 0.725 16 4 algorithms to achieve that good balance between (exploitation)
WOASA 0.255810 0.614 15 138 and (exploration) capabilities of the modified algorithms. The
WOA2SA 0.274970 0.5484 16 142 hybrid memetic algorithms are employed to improve the
performance of general classification tasks and so allow the
WOA3SA 0.260257 0.5809 21 142
hybrid memetic algorithm to predict the terrorist group (s)
70 Iter. WOA 0.283738 0.706 17 6 responsible of the terror attacks on Egypt during the time
WOASA 0.238657 0.685 15 194 period from year 1996 to year 2017 based on the data derived
from the Global Terrorism Database (GTD). The results of the
WOA2SA 0.248016 0.580 21 200
novel memetic hybrid algorithms (WOA2SA), and (WOA3SA)
WOA3SA 0.253790 0.563 17 194 proved the superiority in achieving higher accuracy and
100 Iter. WOA 0.275424 0.681 18 9 maximum fitness than the hybrid original algorithm WOASA
WOASA 0.279005 0.651 23 269 as well as the single WOA which improve the prediction of the
terrorist group(s) responsible of terrorist attacks on Egypt.
WOA2SA 0.278778 0.528 22 267
For future research, the proposed algorithm can be used to
WOA3SA 0.253790 0.5630 17 194 predict the terrorist groups (s) on different regions (countries),
150 Iter. WOA 0.282550 0.718 11 12 a researcher can also use a hybrid embeded features selection
WOASA 0.269197 0.580 18 396
approach instead of wrapper approach that has been used in
our research. A researcher can used different ensemble
WOA2SA 0.239716 0.645 15 466 classification algorithms instead of KNN such as Random
WOA3SA 0.259016 0.539 18 470 forests classification algorithm.
Another interesting research point; a hybridization between
more than two metaheuristics could be applied in our proposed

286
system such as a hybridization between a WOA2 or WOA3 [4] Holland, J.H. (1975). Adaptation in Natural and
with the Particle Swarm Optimization algorithm (PSO) and Artificial Systems. Ann Arbor, MI, USA: Michigan Press
SA. University.
[5] Koza, J.R. (1992). Genetic Programming. Cambridge,
Table 4. The results of the memetic algorithms with 10 USA: MIT Press.
whales [6] Colorni, A., Dorigo, M., Maniezzo, V. (1991).
Distributed optimization by ant colonies. In European
Number Conf. on Artificial Life. Elsevier Publishing, pp. 134–
of Exe. 142.
Mean Selected Time [7] Glover, F. (1977). Heuristics for integer programming
Iteration Alg. Fitness ACC. features (Sec.) using surrogate constraints. Decision Sciences, 8: 156-
5o
iterations WOA 0.30643 0.694 17 2.7
166. https://doi: 10.1111/j.1540-5915.1977.tb01074.x
[8] Mirjalili, S. (2015). Moth-flame optimization algorithm:
WOASA 0.30135 0.575 17 137 a novel nature-inspired heuristic paradigm. Elsevier,
WOA2SA 0.28843 0.527 23 143 Knowledge-based Systems, 89: 228-249.
WOA3SA 0.29472 0.511 20 152 https://doi.org/10.1016/j.knosys.2015.07.006
70 [9] Blum, C., Blesa Aguilera, M.J., Roli, A., Sampels, M.
iterations (2008). Hybrid meta-heuristics – an emerging approach
WOA 0.28374 0.715 24 4 to optimization. Volume 114 of Studies in Computational
Intelligence. Springer.
WOASA 0.28852 0.591 14 230
[10] Krasnogor, N., Smith, J. (2005). A tutorial for competent
WOA2SA 0.275 0.556 22 281 memetic algorithms: Model, taxonomy, and design
WOA3SA 0.29582 0.605 23 382 issues. IEEE Transactions on Evolutionary Computation,
100 9(5): 474488. https://doi:10.1109/TEVC.2005.85020
iterations [11] Ingber, L. (1993). Simulated annealing: Practice versus
WOA 0.29404 0.707 17 8 theory. Mathl. Comput. Modelling, 18(11): 29-57.
https:// doi: 10.1016/0895-7177(93)90204-C
WOASA 0.27411 0.605 21 353
[12] Mirjalili, S., Lewis, A. (2016). The whale optimization
WOA2SA 0.26309 0.527 18 363 algorithm. Advances in Engineering Software, 95: 51–67.
WOA3SA 0.27149 0.621 23 332 https://doi.org/10.1016/j.advengsoft.2016.01.008
150 [13] Soliman, G.M.A., Khorshid, M., Abou-El-Enien, T.H.M.
iterations (2016). Modified moth-flame optimization algorithms
WOA 0.29693 0.704 19 8 for terrorism prediction. International Journal of
WOASA 0.2692 0.595 15 823 Application or Innovation in Engineering &
Management, 5(7): 47-596.
WOA2SA 0.29693 0.625 19 775
[14] Soliman, G.M.A., Abou-El-Enien, T.H.M., Emary, E.,
WOA3SA 0.28821 0.538 19 449 Khorshid, M. (2018). A hybrid whale optimization
algorithm with adaptive spiral for terrorism prediction
(the case of Egypt). European Journal of Scientific
REFERENCES Research, 149(2): 165-184.
[15] Bowser, E.A. (1880). An Elementary Treatise on
[1] Kirkpatrick, S., Gelatt, C.D., Vecchi, M.P. (1983). Analytic Geometry: Embracing Plane Geometry and an
Optimization by simulated annealing. Science, Introduction to Geometry of Three Dimensions (4th ed.).
220(4598): 671–680. New York, D. Van Nostrand.
https://doi:10.1126/science.220.4598.671 [16] Global Terrorism Database (GTD). (2009-2018).
[2] Glover, F. (1989). Tabu Search-Part I. ORSA Journal of University of Maryland. National Consortium for the
Computing, 1(3): 190–206. Study of Terrorism and Responses to Terrorism. A
https://doi.org/10.1287/ijoc.1.3.190 Center of Excellence of the U.S. Department of
[3] Lawler, E.L. (1976). Combinatorial Optimization: Homeland Security. University of Maryland, USA.
Networks and Matroids. Holt, Rinehart and Winston,
USA.

287