Knowledge-Based Systems 197 (2020) 105889

Contents lists available at ScienceDirect

Knowledge-Based Systems
journal homepage: www.elsevier.com/locate/knosys

Fractional-order calculus-based flower pollination algorithm with

local search for global optimization and image segmentation
∗
Dalia Yousri a , , Mohamed Abd Elaziz b , Seyedali Mirjalili c
a
Department of Electrical Engineering, Faculty of Engineering, Fayoum University, Fayoum, Egypt
b
Department of Mathematics, Faculty of Science, Zagazig University, Zagazig, Egypt
c
Centre for Artificial Intelligence Research and Optimisation, Torrens University Australia, Fortitude Valley, Brisbane, 4006 QLD, Australia

article info a b s t r a c t

Article history: Introducing a novel approach to enhance the diversification and intensification propensities of the
Received 31 December 2019 flower pollination algorithm (FPA) is the main aim of this paper. Therefore, the fractional-order (FO)
Received in revised form 6 March 2020 calculus features are adopted to enhance the basic FPA local search ability and adaptive modify the
Accepted 6 April 2020
harmonization coefficient among the FPA exploration and exploitation cores. The proposed Fractional-
Available online 11 April 2020
order FPA (FO-FPA) is examined in a number of experiments. Firstly, FO-FPA is tested with thirty-six
Keywords: benchmark functions with several dimensions. The proposed FO-FPA is compared with recent proved
Flower pollination algorithm techniques based on several statistical measures and non parametric tests. Secondly, FO-FPA is
Local search implemented for a real application of the image segmentation and its results compared with state-
Meta-heuristic of-the-art multi-level thresholding algorithms. The comparisons divulge the remarkable influence of
Image segmentation merging FO with the basic FPA in improving the quality of the solutions and the acceleration of the
Fractional calculus convergence speed.
Optimization
© 2020 Elsevier B.V. All rights reserved.
Algorithm
Benchmark

1. Introduction (diversification and intensification capabilities). Accordingly, the

researches turned for discovering methods to modify these main
In the last two decades, the meta-heuristic (MH) optimization cores of the reported techniques to modify their performance.
algorithms became dominant techniques in several optimization In this context, numerous methods are followed for improving
problems [1]. This motivated researchers to develop a series the behavior of the meta-heuristic algorithms. Some of these
of MH optimization techniques with different inspiration ap- methods included gathering the features of two MH algorithms as
proaches. The MH can be classified for three main categories in [10] where the simulated annealing technique was enhanced
based on their inspiration methodologies namely (I) Evolutionary by the sine-cosine algorithm as well as the moth-flame optimizer
Algorithms (EA), (II) Physics-based algorithms (Ph-A), (III) swarm and differential evolution as in [11]. Merging the chaos maps with
Intelligence (SI) techniques. EA techniques emulate the biological the optimization algorithms to modify their performance was also
introduced as a new approach in [12–14].
evolutionary behaviors such as crossover, mutation, and selection.
In the same context, the flower pollination algorithm (FPA) is
The genetic algorithm (GA) is considered the most popular variant
one of the most popular MH algorithms for several researchers
of this class [2]. For the Ph-A, the physical laws are the main ideas
to be implemented in their applications [15,16]. However the
behind this class of the algorithms such as Henry gas solubility
FPA exhibits good performance for some applications, and it is
optimization (HGSO) [3], Multi-verse optimizer (MVO) [4], and
often failed to find reasonably good solutions in solve prob-
gravitational search algorithm [5]. Finally, the social behaviors lems. Therefore, several modifications for FPA have been devel-
among humans or flock of birds or swarm of animals is the oped to enhance its ability to find a better solution. Some of
inspiration of the third MH class of SI. Particle swarm optimizer them used chaotic maps to improve the FPA. For example, Arora
(PSO) [6], Grey wolf optimizer (GWO) [7], Cuckoo search (CS) [8], et al. [17] applied the chaotic maps to update the parameters to
and Salp swarm Algorithm [9]. As per the literature, several enhance the convergence of FPA and to prevent it from stuck
inspiration approaches were employed while the main cores of on local point. Abdel-Raouf et al. [18] using the chaotic levy
all the algorithms are the exploration and exploitation tendencies flight to improve the performance of the FPA. Whereas the elite
opposition-based technique is applied to the global search strat-
∗ Corresponding author. egy to enhance the ability of the FPA to explore the search space,
E-mail addresses: [email protected] (D. Yousri), meanwhile, the local self-adaptive greedy strategy is used during
[email protected] (M.A. Elaziz), [email protected] (S. Mirjalili). the exploitation stage to guarantee to increase the convergence

https://doi.org/10.1016/j.knosys.2020.105889
0950-7051/© 2020 Elsevier B.V. All rights reserved.
2 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889

Fig. 1. Flowchart of FO-FPA.

towards the global solution. Wang et al. [19] presents the bee (NFL) [33] that states no algorithm is efficient for all optimiza-
pollinator concept in which the operator of the crossover is tion problems. We refer to this theory to justify the need to
used to enhance the diversity and using the elite-based mutation improve FPA and propose its new variant to better solve image
method. Moreover, the fuzzy inference mechanism is used to segmentation problems.
update the probability of switching between the local and global In that circumstance, a novel topic introduced recently which
pollination as described in [20]. Dubey et al. [21] modified the is incorporating an attractive mathematical branch into the opti-
FPA by replacing the random number by a constant scaling factor mization algorithms that named as fractional-order (FO) calculus
to control the local process. to benefit from the memory properties of FO calculus. As FO
Besides, there are other modifications based on combining calculus provides a sufficient description of the memory and the
the FPA with other meta-heuristic methods. Such as Chakraborty hereditary properties of the processes during its combination
et al. [22] used the differential evolution (DE) algorithm, also, the with the evolutionary techniques [34–36]. The fractional-order
developed DE-FPA combines the two local and global strategies derivative encompasses an infinite number of terms treating the
to remove the required of switching probability. Nabil et al. [23] past values, while the integer-order one requires only a finite and
combined the operators of the FPA with the clonal search process a very limited number. Accordingly, the integer-order derivatives
which aims to increase the diversity of the population. Wang [24] are local operators whereas the fractional-order derivative has a
presented a dimension updating technique to improve the local memory of all previous events [35].
search which aims to increase the diversity of the population. To In that aspect, Mousavi et al. [34] adopted FO calculus to
address the memory of fireflies behavior in purpose of mod-
enhance the FPA’s scalability, the normal distribution mutation
ifying the firefly algorithm (FFA) performance as a result the
operator is used which depends on a reduction method over the
fractional-order firefly algorithm (FO-FFA) was reported. FO-FFA
search space as introduced in [25]. The PSO was combined with
was applied to identify the parameters of the more popular
the FPA and applied to constrained optimization problems as
chaotic systems. The results elucidated that accounting terms
in [18]. In the same context, Zhou et al. [26] applied the operators
from the memory of fireflies past behaviors has a significant
of the PSO as a local search to improve the convergence of the
influence in the performance in comparison with other firefly
PFA. Niu et al. [27] applied the local search operators of the wind-
algorithms (FA). Deshmukh et al. [37] utilized FO calculus tool
driven algorithm to enhance the local and global strategies of
for accelerating the convergence of the grey wolf optimizer. Fi-
FPA. The direct search method is used in [28] to improve the
nally, Pires et al. [36] and Couceiro et al. [35] employed FO
performance and the convergence of the PFA which using during calculus to upgrade the particle swarm optimizer (PSO) and the
the global pollination process. Abdel-Raouf et al. [29] applied the Darwinian PSO (DPSO) performance. Integrating the FO calculus
chaotic Harmony search (HS) algorithm to enhance the global with optimization techniques is recently introduced for few al-
convergence of PFA. Kalra and Arora [30] combining the operators gorithms. Moreover, the developed techniques were tested for
of the firefly algorithm (FA) to improve FPA using FA to perform as limited applications. Therefore, merging FO calculus with op-
a local search and determine the current best solution. Then, the timization techniques is a hot research topic that should be
FPA uses a switching condition to using either the global search expanded for several optimization algorithms and applications.
based on FA or the FPA search method. Rohit et al. [31] proposed In this paper, a novel approach is discussed to enhance the
a hybrid between the bat algorithm (BA) and FPA and applied local search of FPA and guarantee a better harmonization among
it for the synthesis of linear antenna arrays. In this algorithm, the FPA exploration and exploitation cores. This approach in-
the BA and FPA are competitive together to prevent trapping cludes merging FO calculus perspective with FPA to capture the
in locally optimal solutions. Rodrigues et al. [32] proposed an previous solutions from memories during the exploitation stage
adaptive version of FPA which has ability to change the setting for modifying the acceleration convergence trend and to find
of its parameter dynamically. more credible solutions. Moreover, the switching probability fac-
All these improvements have demonstrated that MHs require tor among the FPA cores is adaptive updated based on the FO
modification, tuning, or improvement to solve specific problems. calculus coefficients. As such, the Fractional-order flower polli-
This has been logically proved by the No free lunch theorem nation algorithm (FO-FPA) is introduced as a novel variant of
 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889 3

Fig. 2. Mean convergence curves by FPA, and recommended FO-FPA variants of (a) F 1, (b) F 3, (c) F 5, (d) F 7, (e) F 9, and (f) F 11.

FPA. The proposed algorithm is evaluated on thirty-six bench- into different groups. This performed by maximizing the fitness
marks suit with different classifications as twenty-six scalable function called fuzzy entropy and using the histogram of image
uni-modal and multi-modal functions and ten fixed dimensional as input to the proposed FO-FPA.
multi-model functions. Several statistical measures are used to The main contributions of this study are as follows:
appraise the quality of the proposed algorithm versus FPA and the
state-of-the-art algorithms. Moreover the FO-FPA is applied on 1. Apply FO calculus features as a novel approach in MH to
a real application of the image segmentation. The proposed FO- enrich the local search ability of basic FPA via considering
FPA is used to determine the threshold values to split the images the previous behavior of the agents’ memory. Further, to
4 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889

Fig. 3. Mean convergence curves by FPA, and recommended FO-FPA variants of (a) F 13, (b) F 17, (c) F 19, (d) F 21, (e) F 25 and (f) F 26.

guarantee a balance between the FPA local and global 3. Evaluate the proposed FO-FPA to solve real application
search strategies, the switching probability coefficient be- such as image segmentation.
tween these strategies is adaptive tuned based on the FO 4. Compare the performance of the proposed method with
calculus factors. other Meta-heuristic methods to solve global optimiza-
2. Apply the modified FO-FPA as global optimization algo- tion problems, as well as, multi-level thresholding image
rithm to find the optimal solution for thirty-six functions. segmentation problem.
 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889 5

Fig. 4. Tested images and their histograms.

The rest of this paper is organized as follows: in Section 2, the In Section 4, the comparative results between the proposed FO-
basic information about the Flower Pollination Algorithm (FPA) FPA and other MH methods have been performed through two
and the mathematics definition of the Fractional calculus are experimental series (one for global optimization, and the sec-
introduced. Section 3 presents the proposed method according to ond for multi-level image segmentation). Section 5 presents The
improve the performance of FPA using the FO calculus concept. conclusion and future work.
6 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889

Fig. 5. Results of PSNR in terms of (a) Average (b) Rank at each value of r and α .

Fig. 6. Average of the PSNR values overall tested images.

2. Background Yange et al. [38] supposed that the global mating between
plants happens during the biotic process. This case can be math-
2.1. Flower Pollination Algorithm (FPA) overview
ematically modeled as follows:
The primarily inspiration of the FPA technique comes from
the biological concepts of the mating process (pollination) in Zit +1 = Zit + γ L(λ)(g∗ − Zit ), (1)
planets [38]. The pollination process occurs while transmitting
the pollen grains via biotic and/or abiotic operations. In the Biotic where Zit illustrates the solution at tth iteration and ith agent. The
operation, the butterflies, insects carry the pollen grains to a g∗ denotes the global solution obtained. The scaling factor γ is
flower from diverse species; thus, this operation considered a
used to control the step size of levy. The symbol of L(λ) indicates
cross pollination process. Whereas, the abiotic operation is a self
pollination as the wind transmits the pollen grains among flowers the levy factor. Yang et al. [38] used L(λ) to be responsible for
of the same type. transmitting of grains between diverse species of flowers. The
 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889 7

Fig. 7. Average of the SSIM values overall tested images.

Fig. 8. Average of Fitness value for each algorithm.

Fig. 9. Average of CPU time(s) for each algorithm.

levy factor can be calculated as below [38]: - where Zjt , Zkt are the different pollens of same flower species with
index j and k. The symbol of ϵ is withdrawn from a uniform
λΓ (λ) sin( π2λ ) 1 distribution as ϵ ∈ [0, 1]. To switch between the global and local
L∼ (m ≫ mo > 0), mo = 0.1 (2)
π m1+λ mating processes, a switching probability factor S is utilized in
where Γ (λ) represents the gamma function and m represents the the interval of [0.2, 1] [38].
step size.
Yang et al. [38] assumed the pollination via abiotic opera- 2.2. Fractional-order calculus definition
tion can be modeled the local mating process. This approach
represented mathematically as follows: Fractional-order (FO) calculus attracted the interest of numer-
ous researchers due to its flexibility and better fit of the real
Zit +1 = Zit + ϵ (Zjt − Zkt ). (3) responses [39]. FO calculus has several definitions in literature,
8 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889

Table 1
The definition of CEC 2005 Benchmark problems.
ID Formula of function LW UW dimW Type
∑n 2
F1 f (x) = i=1 xi −100 100 30,100,1000 Unimodal
∑n
F2 f (x) = i=1 |xi | + Πin=1 |xi | −10 10 30,100,1000 Unimodal
∑n ∑i 2
F3 f (x) = i=1 ( j−1 xi ) −100 100 30,100,1000 Unimodal

F4 f (x) = maxi {|xi |, 1 ⩽ i ⩽ n} −100 100 30,100,1000 Unimodal

∑n−1
F5 f (x) = i=1 [100(xi+1 − x2i )2 + (xi − 1)2 ] −30 30 30,100,1000 Unimodal
∑n
F6 f (x) = i=1 ( [xi + 0.5]) 2
−100 100 30,100,1000 Unimodal
∑n
F7 f (x) = i=1 ix4i + random[0, 1] − 1.28 1.28 30,100,1000 Unimodal
∑n
F8 f (x) = i=1 ix2i −10 10 30,100,1000 Unimodal
∑n 4
F9 f (x) = i=1 ixi − 1.28 1.28 30,100,1000 Unimodal
∑n i+1
F10 f (x) = i=1 |xi | −1 1 30,100,1000 Unimodal
∑n √
F11 f (x) = i=1 −xi sin( |xi |) −500 500 30,100,1000 Multimodal
∑n
F12 f (x) = [x2i − 10cos(2π xi ) + 10]
i=1 − 5.12 5.12 30,100,1000 Multimodal
√ ∑
n ∑n
F13 f (x) = −20exp(−0.2 1n 2 1
i=1 xi ) − exp( n i=1 cos(2π xi )) + 20 + e −32 32 30,100,1000 Multimodal

∑n x
F14 f (x) = 1
4000 i=1 x2i − Πin=1 cos( √i ) + 1 −600 600 30,100,1000 Multimodal
i

π
∑n−1 ∑n
F15 f (x) = n
{10sin2⎧ (π y1 ) + i=1 (yi − 1)2 [1 + 10sin2 (π yi+1 )] + (yn − 1)2 } + i=1 u(xi , 10, 100, 4) −50 50 30,100,1000 Multimodal
⎨k(xi − a) , xi > a
⎪ m

u(xi , a, k, m) = 0, − a ⩽ xi ⩽ a
⎩k(−x − a)m , x < −a
⎪
i i
∑n ∑n
F16 f (x) = 0.1{sin2 (3π x1 ) + i=1 (xi − 1)2 [1 + sin2 (3π xi + 1)] + (xn − 1)2 [1 + sin2 (2π xn )]} + i=1 u(xi , 5, 100, 4) −50 50 30,100,1000 Multimodal
∑n
i=1 (xi − 1) + [1 + sin (3π xi + 1)] + sin (3π x1 ) + |xn − 1|[1 + sin (3π xn )]
2 2 2 2
F17 f (x) = −10 10 30,100,1000 Multimodal
∑n
F18 f (x) = i=1 |xi sin(xi ) + 0.1xi | −10 10 30,100,1000 Multimodal
∑n ∑n
F19 f (x) = 0.1n − (0.1 i=1 cos(5π xi )) − i=1 x2i −1 1 30,100,1000 Multimodal
∑n 2 ∑n ∑n
i=1 0.5ixi ) + ( i=1 0.5ixi )
2 4
F20 f (x) = i=1 xi + ( −5 10 30,100,1000 Multimodal
√
∑n sin2 ( 100x2i−1 +x2i −0.5)
F21 f (x) = i=1 0.5 + −5 10 30,100,1000 Multimodal
1+.001(x2i −2xi−1 xi +x2i )2

∑n−1
F22 f (x) = 0.1sin2 (3π x1 ) + i=1 (xi − 1)2 (1 + sin2 (3π xi+1 )) + (xn − 1)2 (1 + sin2 (3π xn )) −5 5 30,100,1000 Multimodal

6 (i−1)/(n−1) 2
∑n
F23 f (x) = i=1 (10 ) xi −100 100 30,100,1000 Multimodal
∑n
i=1 (xi − π ) )
f (x) = (−1)n+1 Πin=1 (cos(xi ))exp( 2
F24 −100 100 30,100,1000 Multimodal
√∑ √∑
n n
F25 f (x) = 1 − cos(2π i=1 xi ) + 0.1
2 2
i=1 xi −100 100 30,100,1000 Multimodal
√∑
n
i=1 xi )−0.5
sin2 ( 2
F26 f (x) = 0.5 + ∑n −100 100 30,100,1000 Multimodal
(1+0.001( 2 2
i=1 xi ))

1
∑25 1 −
F27 f (x) = ( 500 + j=1 j+∑2 (x −a )6 ) 1 − 65.536 65.536 2 Multimodal
i=1 i ij

∑11 x1 (b2i +bi x2 ) 2

F28 f (x) = ( i=1 [ai − ] ) −5 5 4 Multimodal
b2i +bi x3 +x4

F29 f (x) = (4x21 − 2.1x41 + 1 6

x
3 1
+ x1 x2 − x22 + 4x42 ) −5 5 2 Multimodal

5.1 5 1
F30 f (x) = (x2 − 4π 2
x21 + x
π 1
− 6)2 + 10(1 − 8π
)cosx1 + 10 −5 5 2 Multimodal

F31 f (x) = [1 + (x1 + x2 + 1)2 (19 − 14x1 + 3x21 − 14x2 + 6x1 x2 + 3x22 )] −2 2 2 Multimodal

×[30 + (2x1 − 3x2 ) × (18 − 32x1 + 12x21 + 48x2 − 36x1 x2 + 27x22 ) ]

∑4 ∑3
F32 f (x) = − i=1 ci exp(− j=1 aij (xj − pij )2 ) 1 3 3 Multimodal

(continued on next page)

 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889 9

Table 1 (continued).
ID Formula of function LW UW dimW Type
∑4 ∑6 2
F33 f (x) = − i=1 ci exp(− j=1 aij (xj − pij ) ) 0 1 6 Multimodal
∑5
F34 f (x) = − i=1 [(X − ai )(X − ai )T + ci ]−1 0 10 4 Multimodal
∑7
F35 f (x) = − i=1 [(X − ai )(X − ai )T + ci ]−1 0 10 4 Multimodal
∑10 T −1
F36 f (x) = − i=1 [(X − ai )(X − ai ) + ci ] 0 10 4 Multimodal

Grunwald–Letnikov is one of these definitions, and it can be reformulated to match the definition in Eq. (7) for α = 1 as
mathematically formulated as below [40]: follow.
D1 [Z (t + 1)] = Zit +1 − Zit = ϵ Zjt − Zkt .
∞
( )
α (8)
( )
α 1 ∑
D (x(t)) = lim α (−1)n x(t − nh), (4)
h→0 h n By substituting the plain derivative to the fractional-order one
n=0
based on the GL general definition with derivative orders (α ), we
where
can have the following relation:
α Γ (α + 1) α (α − 1)(α − 2)...(α − n + 1)
( )
Dα Zit +1 = ϵ Zjt − Zkt .
[ ] ( )
= = , (9)
n Γ (n + 1)Γ (α − n + 1) n!
By using GL definition in Eq. (6), the expression of Eq. (9) for
(5) T = 1 can be modeled as follow:
where, Dα (x(t)) is the Grunwald–Letnikov (GL) fractional deriva- r
(−1)n Γ (α + 1)Zit +1−n
tive of order α . The symbol of Γ indicates the Euler gamma Dα Zit +1 = Zit +1 +
∑
= ϵ Zjt − Zkt .
[ ] ( )
function. Γ (n + 1)Γ (α − n + 1)
n=1
In discrete-time implementation, the GL definition for a signal
(10)
(x(t)) in Eq. (4) can be formulated as follows [41]:
r According for the previous relation in Eq. (10), the general ex-
1 ∑ (−1)n Γ (α + 1)x(t − nT )
Dα [x(t)] = (6) pression for the solutions update of FO-FPA can be formulated as
Tα Γ (n + 1)Γ (α − n + 1) below:
n=0
r
where T is the sampling period and r is the number of terms from
∑ (−1)n Γ (α + 1)Zit +1−n
Zit +1 = − + ϵ Zjt − Zkt .
( )
(11)
previous events or memory. The x(t) is considered a discrete- Γ (n + 1)Γ (α − n + 1)
n=1
variable bounded real-valued function
For the special case of α = 1, Eq. (6) can be rewritten as below: For considering the first four terms (r = 4) from the historical
data with derivative order α , the solutions of FO-FPA is updated
as follow;
D1 [x(t)] = x(t + 1) − x(t) (7)
1 1 1
Zit +1 = α Zit + α (1 − α )Zit −1 + α (1 − α )(2 − α )Zit −2
where D1 [x(t)] is the difference between two followed events 1! 2! 3! (12)
(plain derivative). 1
α (1 − α )(2 − α )(3 − α )Zit −3 + ϵ Zjt − Zkt ,
( )
+
4!
3. Proposed Fractional-order Flower Pollination Algorithm For the first 8 terms of the historical data, the solutions of FO-FPA
(FO-FPA) is modified as below;

In this part, a novel approach is introduced to enhance the 1 1 1

Zit +1 = α Zit + α (1 − α )Zit −1 + α (1 − α )(2 − α )Zit −2 +
local search of the classic version of FPA to control its acceleration 1! 2! 3!
convergence trend and achieve creditable solutions. In addition 1
α (1 − α )(2 − α )(3 − α )Zit −3
an adaptive tuning for the switching probability factor among 4!
FPA’s exploration and exploitation cores is addressed. This ap- 1
proach presents merging the fractional calculus with the basic + α (1 − α )(2 − α )(3 − α )(4 − α )Zit −4 +
5!
FPA as a result Fractional-order FPA is generated (FO-FPA). The 1
details of the introduced algorithm are listed in the following α (1 − α )(2 − α )(3 − α )(4 − α )(5 − α )Zit −5 +
6!
subsections. 1
α (1 − α )(2 − α )(3 − α )(4 − α )(5 − α )(6 − α )Zit −6 +
7!
3.1. Enhancement the local search of FPA based on FO calculus 1
α (1 − α )(2 − α )(3 − α )(4 − α )(5 − α )(6 − α )(7 − α )Zit −7
8!
FPA shows a good global search inclination, but it suffers from
+ ϵ Zjt − Zkt ,
( )
the inefficient local search that led to a poor decaying in the con-
vergence speed. In the purpose of tackling this limitation, in this (13)
approach, the FO calculus embedded into the basic FPA to benefit
from the FO calculus memory property of the previous solutions 3.2. Adaptive tuning of switching probability factor (S) based on FO
to guarantee sharing the information among the solutions in the calculus
exploitation phase. Accordingly the accuracy of the solutions and
the convergence rate are modified. Achieving a balance between the global and local search
To apply the early mentioned definition of FO calculus to propensities of the optimization algorithms is a major aim to
boost the FPA local search of Eq. (3), first, the Eq. (3) can be converge for the optimal solutions. In the FO-FPA, the switching
10 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889

probability S is adaptive withdrawn from the B distribution. The The pseudo code of the introduced FO-FPA technique is reported
B function is calculated for the derivative order vector (α ) from in Algorithm 1.
0.1 to 1 with step of 0.1 and number of terms r equal 4, or 8 or
12 via the relation in (14b) as follow; Algorithm 1 Pseudo-code of FO-FPA algorithm
y = B((αv ec ), r), (14a) Inputs: The population size N, maximum number of iterations
(y − min(y)) ∗ (b − a) T , derivative order coefficient α and number of terms r. Nor-
S(j) = +a (14b) malized vector (y) of a B function for the derivative orders
(max(y) − min(y))
vector and number of terms r Eq. (14a).
where B is the beta function with inputs of the vector of the Outputs: The best solution and the accomplished fitness
derivative order (α ) from 0.1 to 1 with step of 0.1 and number of function.
terms r of 4, or 8 or 12. The generated distribution y has minimum Initialize the random population of N flowers/pollen.
and maximum values can be given by min(y) and max(y). The Calculate the fitness values.
normalization interval ranges are a and b where they equaled to
Determine the global solution g∗ .
0.9 and 0.6, respectively. The j is the index of the used derivative
Initialize the past solutions (memory property) .
order j = 1, 2, 3,.., 10 for α = 0.1, 0.2, 0.3, . . . , 1.
Calculate S based on r and α via Eq. (14b) .
while (t< T (Stop criteria is not satisfied)) do
3.3. Implementation steps of FO-FPA
for (i = 1 : N (all N followers in the population)) do .
if (rand > S) then
The structure of the FO-FPA is given in Fig. 1. In general, the
Do Global Search process based on Eq. (1)
proposed FO-FPA consists of three stages which can be summa-
else
rized as follows;
Do local Search process based on Eq. (11) based on
• Initialization stage: FO-FPA starts by set of random solutions derivative order α and terms r
for a set of N flowers. Based on these initial solutions (Zil ), End if
the initial fitness function values and the global solution Update the past solutions (memory terms (r)) on FIFO
obtained so far (g∗ ) are assigned. Moreover, the previous bases .
events for the initialization state are determined as a histor- Compute the fitness function fi (i = 1, 2, . . . , N).
ical stored data early (Zoldil ). For this purpose the following End for
formula is performed; Retain the best solution g∗ .
End While
Zil = LBil + rand() × (UBil − LBil ), i = 1, 2, . . . , N , l = 1, 2, . . . , D.
Return the best solution
(15)
In Eq. (15), UBil and LBil represents the boundaries for agent i 3.4. Computational complexity of FPA and FO-FPA algorithms
at dimension l. rand() is a uniform random distribution that
occupies numbers between 0 and 1. The Computational Complexity of the algorithm is controlled
• Update the solution stage: in this stage, the initial solutions by its structure as Computational complexity is a pointer of its
are updated until the termination criteria is met. In the main execution time. The computational complexity of the FPA bases
structure of the proposed algorithm, the transition process on the number of flowers/pollen (N), dimension of the problem
among the global and local search processes is controlled by (d) and the maximum number of iterations (T ), whereas for the
a switching probability factor (S). In FO-FPA, the FO calculus FO-FPA, the number of the stored terms of memory (r) is also
factors are utilized to compute the corresponded value of accounted in computing its computational time. Accordingly, the
S as mention in Section 3.2. For each agent, the solutions total computational complexity of FPA is O(N × (d + T )) while of
are updated based on Eq. (1) of a global search process in FO-FPA is O(N × (d × r + T + T × r)). Where the computational
case of rand > S, otherwise the solutions are updated based complexity of FPA for the initialization process can be calculated
on the enhance local search process by GL definition where as O(N × d), whereas FO-FPA is O(N × d × r). The computational
n time of the main execution loop of FPA is O(T × N), while that of
the previous solutions from memory (Zold , where n changes
ij
from 1 to r) as illustrated in Eq. (11). The historical events FO-FPA is (T × N × r + T × N).
from memory are accounted to guarantee sharing the earlier
solutions to enhance the local search tendency. Each of these 4. Numerical validation
( 1 ) (in1 its weight) (ω
solutions multiplied ( 11 to ωr ) that appeared ) in In this study, to compute the quality of FO-FPA, we conducted
Eq. (11) as 1!
α , 2!
α (1 − α ) , 3 !
α (1 − α )(2 − α ) ,....
two experiments. In the first experiment, the efficiency of the
α (1 − α )......((r − 1) − α ) (see Fig. 1). At the end of each
(1 )
r! FO-FPA as global optimization is assessed. Meanwhile, the second
iteration in this stage, the old solutions are updated based experiments aims to apply the FO-FPA to a real image segmenta-
on the concept of first in first out (FIFO) where the historical tion applications. The details of the these two experimental series
solutions of each agent are updated by using the recent so- are given in the following two subsections.
lution before the next iteration as clarified from Fig. 1. FIFO
is a well-known concept in for organizing and manipulating 4.1. Experimental series 1: Global optimization
a data where the oldest solution saved firstly then after
modification; it becomes not effective therefore its value is The quality of the introduced FO-FPA variants is assessed ver-
modified with the later solution as illustrated in Fig. 1 as sus the classical version of FPA via testing with 36 well-regarded
below. benchmark functions of CEC 2005 [42]. The considered function
• Termination criteria stage: this stage is responsible for stop- are categorized as uni-modal, multi-modal functions, and fixed-
ping the execution of the proposed algorithm. In the current dimensional multi-model benchmarks. The function suite in-
work satisfying the maximum number of iterations (T) is the cludes twenty-six scalable problems and nine fixed-dimensional
terminal condition.
Table 2
Comparison of simulation results of FPA and FO-FPA algorithms with number of terms 4 and α changed from 0.1 to 1 for uni-model and multi-model functions with dimension 30 as well as for fixed-dimensional
multi-modal functions.
Algorithms
Functions FPA FO-FPAα=0.1 FO-FPAα=0.2 FO-FPAα=0.3 FO-FPAα=0.4 FO-FPAα=0.5 FO-FPAα=0.6 FO-FPAα=0.7 FO-FPAα=0.8 FO-FPAα=0.9 FO-FPAα=1
Mean 2.0542e+03 3.7049e−249 2.4515e−164 3.3471e−128 3.4825e−103 4.1363e−77 1.2410e−55 2.7684e−34 2.3041e−17 2.6046e−04 1.5540e+02
STD 7.2876e+02 0.0000e+00 0.0000e+00 1.1304e−127 9.7544e−103 1.4061e−76 1.5431e−55 1.0816e−33 2.0712e−17 1.9859e−04 8.0802e+01
F1
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 2.9404e+01 4.8585e−126 3.1539e−83 4.1678e−65 2.1191e−52 4.6890e−40 2.9268e−28 5.8147e−18 3.1345e−09 1.5045e−02 1.5455e+01
STD 4.9789e+00 7.5972e−126 7.9101e−83 7.3365e−65 4.6439e−52 5.6904e−40 2.8984e−28 5.4754e−18 1.4744e−09 8.1390e−03 7.2603e+00
F2
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 1.4789e+03 3.5784e−248 5.2716e−163 1.3535e−126 9.4691e−102 8.8403e−78 2.1256e−54 2.9677e−33 8.3279e−17 6.0664e−04 1.7412e+02

D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889
STD 4.0380e+02 0.0000e+00 2.2228e−162 7.3293e−126 2.4941e−101 1.5973e−77 4.2037e−54 8.9751e−33 9.7388e−17 3.7930e−04 8.4145e+01
F3
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 2.3949e+01 3.9135e−125 5.4986e−82 1.6232e−64 7.7773e−52 3.3530e−39 1.3534e−27 3.5642e−17 1.8730e−08 3.8283e−02 8.2973e+00
STD 2.8292e+00 6.8126e−125 9.1569e−82 3.5705e−64 1.5429e−51 6.1912e−39 1.7615e−27 3.6527e−17 1.0343e−08 1.3054e−02 2.3664e+00
F4
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 3.3292e+05 2.8950e+01 2.8924e+01 2.8919e+01 2.8916e+01 2.8891e+01 2.8898e+01 2.8898e+01 2.8892e+01 2.8911e+01 4.5179e+03
STD 1.5873e+05 2.3685e−02 2.1889e−02 1.8841e−02 2.5727e−02 5.2572e−02 4.3576e−02 4.2380e−02 4.9761e−02 3.7601e−02 3.6443e+03
F5
p-value 3.0199e−11 4.3106e−08 0.0019527 0.018368 0.059428 + 0.66273 0.86499 0.54933 0.20095 3.0199e−11
Rank 11 9 8 7 6 1 3 4 2 5 10
Mean 1.7825e+03 6.2876e+00 5.5717e+00 5.6319e+00 5.4737e+00 5.3924e+00 5.2665e+00 5.2114e+00 4.5023e+00 4.4830e+00 1.4817e+02
STD 4.6470e+02 5.5802e−01 7.7128e−01 5.4997e−01 5.0728e−01 5.1698e−01 5.4983e−01 4.1745e−01 7.0468e−01 6.1549e−01 8.4975e+01
F6
p-value 3.0199e−11 1.4643e−10 7.043e−07 1.85e−08 7.695e−08 4.444e−07 1.3367e−05 1.4918e−06 0.95873 + 3.0199e−11
Rank 11 9 7 8 6 5 4 3 2 1 10
Mean 3.5887e−01 7.5874e−05 1.1211e−04 1.1011e−04 1.5650e−04 1.5276e−04 3.2178e−04 5.4931e−04 1.4605e−03 6.1646e−03 4.5461e−02
STD 1.4421e−01 6.6030e−05 1.0409e−04 9.7449e−05 9.9282e−05 1.1838e−04 1.8776e−04 3.0172e−04 7.9063e−04 2.2968e−03 2.5948e−02
F7
p-value 3.0199e−11 + 0.15798 0.16687 0.00085641 0.0036709 1.5581e−08 2.6099e−10 3.3384e−11 3.0199e−11 3.0199e−11
Rank 11 1 3 2 5 4 6 7 8 9 10
Mean 2.5132e+02 4.0208e−250 9.3202e−165 8.6626e−129 8.9094e−104 2.3440e−78 1.1761e−55 3.3851e−35 3.8197e−18 2.4986e−05 2.0261e+01
STD 6.8198e+01 0.0000e+00 0.0000e+00 2.4927e−128 2.8933e−103 5.6321e−78 2.9235e−55 8.5596e−35 3.6977e−18 1.4854e−05 1.1942e+01
F8
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 1.4071e−01 0.0000e+00 0.0000e+00 1.0256e−260 6.6245e−211 1.3977e−158 5.2620e−116 2.1002e−74 3.3983e−40 7.3548e−15 1.5427e−03
STD 8.2755e−02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 7.6552e−158 1.8531e−115 8.0323e−74 1.0519e−39 1.1317e−14 2.2029e−03
F9
p-value 1.2118e−12 + NaN 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12
Rank 10 1 1 2 3 4 5 6 7 8 9
Mean 6.8339e−04 1.4354e−157 1.4717e−110 2.2123e−92 6.5242e−78 4.3603e−64 4.0156e−50 1.8322e−37 7.6605e−25 1.3743e−12 6.9058e−05
STD 5.4258e−04 2.7994e−157 4.1092e−110 9.9960e−92 2.7963e−77 8.6789e−64 9.5608e−50 2.6889e−37 9.7987e−25 2.3609e−12 6.3644e−05
F10
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean −6.3155e+03 −5.0939e+03 −6.0882e+03 −6.3499e+03 −6.4522e+03 −6.4965e+03 −6.5173e+03 −6.7749e+03 −6.9323e+03 −7.0587e+03 −7.0780e+03
STD 3.4840e+02 3.7571e+02 2.9504e+02 3.2448e+02 3.2557e+02 2.5592e+02 2.9670e+02 2.6455e+02 1.9490e+02 2.3697e+02 2.3238e+02
F11
p-value 1.6947e−09 3.0199e−11 4.0772e−11 1.287e−09 1.0702e−09 8.891e−10 1.3111e−08 6.356e−05 0.012212 0.69522 +
Rank 9 11 10 8 7 6 5 4 3 2 1
Mean 1.8594e+02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 2.7386e+01 8.8797e+01 1.3543e+02
STD 1.5546e+01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 2.3181e+01 9.5494e+00 1.7506e+01
F12
p-value 1.2118e−12 + NaN NaN NaN NaN NaN NaN 1.2118e−12 1.2118e−12 1.2118e−12
Rank 5 1 1 1 1 1 1 1 2 3 4

(continued on next page)

11
 12
Table 2 (continued).
Algorithms
Functions FPA FO-FPAα=0.1 FO-FPAα=0.2 FO-FPAα=0.3 FO-FPAα=0.4 FO-FPAα=0.5 FO-FPAα=0.6 FO-FPAα=0.7 FO-FPAα=0.8 FO-FPAα=0.9 FO-FPAα=1
Mean 7.8934e+00 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 1.2799e−09 4.6141e−03 5.3107e+00
STD 1.3245e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 6.8830e−10 1.7079e−03 1.4381e+00
F13
p-value 1.2118e−12 + NaN NaN NaN NaN NaN NaN 1.2118e−12 1.2118e−12 1.2118e−12
Rank 5 1 1 1 1 1 1 1 2 3 4
Mean 1.7391e+01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 3.7007e−18 9.0465e−04 2.4515e+00
STD 5.3572e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 2.0270e−17 9.7199e−04 6.9714e−01
F14
p-value 1.2118e−12 + NaN NaN NaN NaN NaN NaN 0.33371 1.2118e−12 1.2118e−12
Rank 5 1 1 1 1 1 1 1 2 3 4
Mean 3.2574e+02 8.6104e−01 7.0622e−01 5.9498e−01 5.6543e−01 5.9506e−01 5.1892e−01 5.3333e−01 4.7991e−01 4.5513e−01 4.0578e+00
STD 1.0779e+03 1.8857e−01 1.2033e−01 9.7420e−02 1.1432e−01 1.4792e−01 1.0781e−01 1.3488e−01 8.6703e−02 1.0986e−01 1.3197e+00

D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889
F15
p-value 3.0199e−11 1.5465e−09 1.1023e−08 5.0912e−06 0.00049818 0.00018916 0.032651 0.021506 0.43764 + 3.0199e−11
Rank 11 9 8 6 5 7 3 4 2 1 10
Mean 1.2086e+05 2.9797e+00 2.9441e+00 2.9622e+00 2.9121e+00 2.9650e+00 2.8150e+00 2.5212e+00 2.4832e+00 2.5031e+00 1.3276e+01
STD 1.4077e+05 5.4459e−02 1.2040e−01 1.0911e−01 2.3830e−01 1.2670e−01 3.1253e−01 3.9861e−01 2.3137e−01 2.9691e−01 6.2333e+00
F16
p-value 3.0199e−11 5.4941e−11 6.121e−10 2.6099e−10 1.2023e−08 2.8716e−10 1.1674e−05 0.65204 + 0.7394 3.0199e−11
Rank 11 9 6 7 5 8 4 3 1 2 10
Mean 4.5656e+01 2.9762e+01 2.9551e+01 2.9565e+01 2.9513e+01 2.9102e+01 2.8738e+01 2.5170e+01 2.4101e+01 2.2107e+01 4.7765e+01
STD 1.5170e+01 7.2434e−01 1.8731e+00 1.4546e+00 1.0967e+00 2.5460e+00 2.7054e+00 3.7730e+00 2.8798e+00 2.5602e+00 9.5557e+00
F17
p-value 3.6897e−11 4.0772e−11 3.8202e−10 9.9186e−11 7.3891e−11 1.6947e−09 6.5183e−09 0.00012477 0.0066689 + 3.0199e−11
Rank 10 9 7 8 6 5 4 3 2 1 11
Mean 2.0866e+01 3.5975e−126 5.9879e−84 9.1108e−67 1.1491e−53 4.1906e−41 7.5319e−29 6.3550e−19 7.7722e−10 8.1208e−02 1.4634e+01
STD 2.3996e+00 1.2186e−125 1.0951e−83 1.1612e−66 2.1965e−53 6.5819e−41 2.5839e−28 4.8683e−19 4.3033e−10 1.0429e−01 3.7497e+00
F18
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 2.1948e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 9.2279e−07 6.8535e−01
STD 3.0675e−01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 7.9967e−07 3.6658e−01
F19
p-value 1.2118e−12 + NaN NaN NaN NaN NaN NaN NaN 1.2118e−12 1.2118e−12
Rank 4 1 1 1 1 1 1 1 1 2 3
Mean 1.2083e+02 7.8105e−250 2.3580e−163 1.8638e−128 1.4043e−102 7.8697e−78 7.6857e−55 5.7809e−34 4.8385e−13 7.4207e−04 6.6494e+00
STD 3.0167e+01 0.0000e+00 0.0000e+00 5.3570e−128 6.6095e−102 2.3999e−77 1.4112e−54 1.6440e−33 2.6265e−12 9.9409e−04 3.4184e+00
F20
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean −1.7399e+01 −1.8414e+01 −2.2024e+01 −2.2194e+01 −2.2517e+01 −2.1875e+01 −2.1175e+01 −1.6383e+01 −1.2950e+01 −1.2235e+01 −1.3765e+01
STD 4.3843e+00 8.4125e−01 8.7950e−01 9.4211e−01 9.7093e−01 9.5337e−01 1.1812e+00 2.9635e+00 3.0066e+00 3.6417e+00 3.9962e+00
F21
p-value 2.7829e−07 3.0199e−11 0.19579 0.25805 + 0.013832 9.2113e−05 3.3384e−11 3.0199e−11 8.9934e−11 1.4643e−10
Rank 7 6 3 2 1 4 5 8 10 11 9
Mean 2.0152e+01 2.9608e+01 2.9695e+01 2.9121e+01 2.8986e+01 2.8741e+01 2.7195e+01 2.4661e+01 2.3945e+01 1.9942e+01 2.8689e+01
STD 5.1967e+00 1.1510e+00 6.5698e−01 2.3926e+00 2.5867e+00 2.3139e+00 3.4972e+00 3.5258e+00 2.6159e+00 3.7916e+00 4.8024e+00
F22
p-value 0.83026 5.4941e−11 3.0199e−11 3.1589e−10 5.0723e−10 5.5727e−10 6.5277e−08 4.084e−05 9.2113e−05 + 2.3897e−08
Rank 2 10 11 9 8 7 5 4 3 1 6
Mean 7.5896e+06 2.4465e−244 1.2825e−158 3.3975e−124 6.9895e−99 2.0795e−74 3.1555e−51 2.4584e−30 4.5083e−13 2.0122e+00 1.2331e+06
STD 2.7360e+06 0.0000e+00 6.4265e−158 8.0928e−124 1.8276e−98 4.9826e−74 6.3304e−51 9.7823e−30 5.6456e−13 1.3179e+00 6.6715e+05
F23
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 0.0000e+00 −2.1138e−100 −2.0466e−92 −1.6721e−89 −1.4544e−82 −2.2019e−86 −1.4296e−86 −3.1687e−77 −4.0388e−179 0.0000e+00 0.0000e+00
STD 0.0000e+00 1.1578e−99 1.1174e−91 9.1537e−89 7.9662e−82 1.1978e−85 7.8295e−86 1.7356e−76 0.0000e+00 0.0000e+00 0.0000e+00
F24
p-value 1.2118e−12 4.4205e−06 0.051877 0.44642 0.02236 0.099258 0.42039 + 1.7203e−12 1.2118e−12 1.2118e−12
Rank 9 7 6 5 2 3 4 1 8 10 11

(continued on next page)

Table 2 (continued).
Algorithms
Functions FPA FO-FPAα=0.1 FO-FPAα=0.2 FO-FPAα=0.3 FO-FPAα=0.4 FO-FPAα=0.5 FO-FPAα=0.6 FO-FPAα=0.7 FO-FPAα=0.8 FO-FPAα=0.9 FO-FPAα=1
Mean 5.5541e+00 2.3641e−119 3.2297e−66 4.0447e−48 2.0855e−28 1.6078e−02 8.9858e−02 9.9873e−02 9.9873e−02 2.3373e−01 2.2914e+00
STD 5.8876e−01 8.6643e−119 1.6897e−65 1.2533e−47 7.9941e−28 1.5922e−02 2.1111e−02 9.4486e−11 1.2188e−09 5.7035e−02 5.9244e−01
F25
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 4.9261e−01 0.0000e+00 1.0443e−03 2.0902e−03 5.8442e−03 9.4632e−03 9.6966e−03 9.7159e−03 9.7159e−03 5.8561e−02 4.7608e−01
STD 4.8227e−03 0.0000e+00 1.4809e−03 2.6774e−03 2.7421e−03 1.1868e−03 1.0595e−04 4.7947e−11 1.8026e−10 2.4566e−02 1.9436e−02
F26
p-value 1.2118e−12 + 1.9457e−09 6.247e−10 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 9.9829e−01 7.6988e+00 6.2151e+00 6.4502e+00 4.5191e+00 2.3145e+00 2.0288e+00 2.3472e+00 1.5141e+00 1.4572e+00 1.0338e+00
STD 1.4466e−03 3.8066e+00 3.5173e+00 2.4608e+00 2.2191e+00 1.1312e+00 9.4024e−01 9.7974e−01 7.6965e−01 6.7084e−01 1.3736e−01
F27
3.3384e−11 3.3384e−11 3.3384e−11 3.6897e−11 8.1527e−11 1.3289e−10 3.6897e−11 1.5465e−09 4.9752e−11 4.3106e−08

D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889
p-value +
Rank 1 11 9 10 8 6 5 7 4 3 2
Mean 6.6576e−04 2.7212e−03 1.0089e−03 9.3299e−04 8.0221e−04 7.0470e−04 8.2110e−04 7.2421e−04 7.3107e−04 7.2439e−04 8.2504e−04
STD 1.4943e−04 2.6418e−03 4.8054e−04 4.2639e−04 2.7953e−04 1.6341e−04 2.4092e−04 1.3593e−04 1.0737e−04 1.4087e−04 1.4188e−04
F28
p-value + 3.8202e−10 0.0027548 0.045146 0.11882 0.55923 0.0098834 0.25188 0.10869 0.24581 0.00017836
Rank 1 11 10 9 6 2 7 3 5 4 8
Mean −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00
STD 9.1095e−08 1.3718e−05 3.4055e−07 2.1731e−07 2.4631e−07 1.7626e−07 2.3317e−07 2.2702e−07 3.5407e−07 2.0834e−07 1.3274e−08
F29
p-value 5.462e−06 8.1527e−11 2.3715e−10 8.1527e−11 1.2057e−10 2.8716e−10 9.9186e−11 8.9934e−11 1.6132e−10 3.8202e−10 +
Rank 2 11 10 6 5 3 9 7 8 4 1
Mean 3.9789e−01 9.5018e−01 5.9592e−01 5.4303e−01 5.1533e−01 5.3890e−01 4.9532e−01 4.9642e−01 4.5369e−01 4.0514e−01 3.9789e−01
STD 5.8700e−12 6.0685e−01 1.6363e−01 1.7441e−01 9.9049e−02 1.6133e−01 1.1933e−01 1.9752e−01 9.4518e−02 6.3723e−03 1.8337e−08
F30
p-value + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.3384e−11
Rank 1 11 10 9 7 8 5 6 4 3 2
Mean 3.0000e+00 1.4030e+01 1.0982e+01 8.2625e+00 7.2208e+00 8.2194e+00 6.0705e+00 4.0740e+00 3.7602e+00 3.1259e+00 3.0000e+00
STD 3.1311e−09 9.6329e+00 8.2482e+00 6.3534e+00 6.6835e+00 7.4626e+00 4.3182e+00 1.3314e+00 1.0416e+00 1.0789e−01 2.8151e−05
F31
p-value + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 1 11 10 9 7 8 6 5 4 3 2
Mean −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01
STD 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16
F32
p-value + NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN
Rank 1 1 1 1 1 1 1 1 1 1 1
Mean −3.2960e+00 −2.3752e+00 −2.8855e+00 −3.0078e+00 −3.0795e+00 −3.0454e+00 −3.0594e+00 −3.0500e+00 −3.0699e+00 −3.0347e+00 −3.0569e+00
STD 1.9661e−02 3.7069e−01 3.0234e−01 1.8259e−01 1.2738e−01 2.0650e−01 1.8690e−01 2.0147e−01 1.6837e−01 1.9876e−01 2.0228e−01
F33
p-value + 3.0199e−11 3.3384e−11 3.6897e−11 3.3384e−11 3.6897e−11 3.3384e−11 3.3384e−11 3.3384e−11 3.0199e−11 6.6955e−11
Rank 1 11 10 9 2 7 4 6 3 8 5
Mean −1.0146e+01 −3.7489e+00 −4.7878e+00 −5.1255e+00 −5.5798e+00 −5.4642e+00 −5.4976e+00 −5.5449e+00 −5.7675e+00 −5.4772e+00 −5.7790e+00
STD 8.1113e−03 1.4643e+00 2.2030e+00 1.8141e+00 2.1964e+00 2.4082e+00 2.4359e+00 2.3861e+00 2.0061e+00 2.2589e+00 2.3739e+00
F34
p-value + 3.0199e−11 3.0199e−11 3.0199e−11 3.3384e−11 3.0199e−11 3.0199e−11 3.3384e−11 3.0199e−11 3.0199e−11 5.4941e−11
Rank 1 11 10 9 4 8 6 5 3 7 2
Mean −1.0309e+01 −3.6123e+00 −4.4695e+00 −4.7396e+00 −5.1513e+00 −5.3543e+00 −5.2336e+00 −5.6658e+00 −5.8741e+00 −4.7289e+00 −5.1905e+00
STD 1.3954e−01 1.0853e+00 1.4401e+00 1.0442e+00 1.2879e+00 1.6394e+00 1.2200e+00 1.5044e+00 2.1064e+00 1.3564e+00 1.6491e+00
F35
p-value + 3.0199e−11 9.9186e−11 3.0199e−11 3.0199e−11 1.0937e−10 3.0199e−11 3.6897e−11 2.1544e−10 3.0199e−11 3.3384e−11
Rank 1 11 10 8 7 4 5 3 2 9 6
Mean −1.0399e+01 −3.9003e+00 −4.7097e+00 −5.1825e+00 −5.3859e+00 −5.8745e+00 −6.2054e+00 −5.7980e+00 −6.1284e+00 −5.7172e+00 −5.3947e+00
STD 1.5282e − 01 1.1178e+00 1.1439e+00 1.0691e+00 1.3815e+00 1.8042e+00 1.9643e+00 1.2424e+00 1.8716e+00 1.7247e+00 1.7863e+00
F36
p-value + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 6.0658e−11 1.3289e−10 3.0199e−11 3.8202e−10 1.4643e−10 4.0772e−11
Rank 1 11 10 9 8 4 2 5 3 6 7
+/= /− 9/ 1/ 26 18/ 0/ 18 6/ 3/ 27 5/ 3/ 28 6/ 2/ 28 6/ 2/ 28 5/ 2/ 29 6/ 3/ 27 3/ 5/ 28 5/ 4/ 27 3/ 0/ 33
Average rank 7.0278 5.4444 5.1667 5 4.3611 4.5556 4.6667 4.8889 5 5.8056 7.1667

13
 14
Table 3
Comparison of simulation results of FPA and FO-FPA algorithms with number of terms 8 and α changed from 0.1 to 1 for uni-model and multi-model functions with dimension 30 as well as for fixed-dimensional
multi-modal functions.
Algorithms
Functions FPA FO-FPAα=0.1 FO-FPAα=0.2 FO-FPAα=0.3 FO-FPAα=0.4 FO-FPAα=0.5 FO-FPAα=0.6 FO-FPAα=0.7 FO-FPAα=0.8 FO-FPAα=0.9 FO-FPAα=1
Mean 2.0542e+03 1.5414e−184 8.8758e−141 1.0147e−118 8.1433e−98 3.5988e−77 1.7997e−54 2.6344e−34 3.8695e−17 2.4177e−04 1.3548e+02
STD 7.2876e+02 0.0000e+00 2.3579e−140 2.7456e−118 4.2479e−97 1.0139e−76 4.3778e−54 4.7315e−34 3.3261e−17 1.7249e−04 6.8131e+01
F1
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 2.9404e+01 5.0414e−93 5.1770e−71 3.5731e−60 8.5075e−51 2.0773e−39 5.5838e−28 7.1966e−18 4.0812e−09 1.5588e−02 1.4460e+01
STD 4.9789e+00 3.4723e−93 1.0465e−70 8.8416e−60 1.1051e−50 2.2659e−39 5.7837e−28 6.5226e−18 2.8039e−09 5.6337e−03 6.8636e+00
F2
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 1.4789e+03 1.2358e−183 1.5451e−139 7.2499e−118 4.8608e−98 7.4196e−76 1.3645e−53 1.5490e−33 1.8812e−16 6.6930e−04 2.1095e+02

D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889
STD 4.0380e+02 0.0000e+00 5.1660e−139 2.0817e−117 1.0969e−97 2.5033e−75 3.7435e−53 2.7534e−33 1.9655e−16 7.8656e−04 9.3169e+01
F3
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 2.3949e+01 9.9764e−93 9.3287e−71 4.1606e−60 9.6543e−50 1.3462e−38 2.0495e−27 3.6448e−17 1.7475e−08 4.3849e−02 9.1728e+00
STD 2.8292e+00 7.3154e−93 1.5146e−70 4.3411e−60 1.6940e−49 1.8078e−38 3.0299e−27 3.2575e−17 9.5672e−09 2.0547e−02 2.4163e+00
F4
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 3.3292e+05 2.8942e+01 2.8921e+01 2.8916e+01 2.8914e+01 2.8901e+01 2.8887e+01 2.8899e+01 2.8887e+01 2.8911e+01 4.2979e+03
STD 1.5873e+05 1.7276e−02 2.4151e−02 3.0532e−02 2.4481e−02 2.2467e−02 4.2292e−02 3.0702e−02 3.6798e−02 3.0454e−02 3.1213e+03
F5
p-value 3.0199e−11 4.1825e−09 0.00018916 0.0013703 0.0030339 0.14128 0.68432 0.20095 + 0.019112 3.0199e−11
Rank 11 9 8 7 6 4 2 3 1 5 10
Mean 1.7825e+03 5.8844e+00 5.5247e+00 5.3460e+00 5.3438e+00 5.2792e+00 5.3419e+00 5.1389e+00 4.8685e+00 4.2273e+00 1.5579e+02
STD 4.6470e+02 5.8687e−01 5.8300e−01 4.1227e−01 4.8928e−01 5.4298e−01 4.6995e−01 4.9964e−01 6.1558e−01 6.7795e−01 7.0097e+01
F6
p-value 3.0199e−11 8.1014e−10 2.3897e−08 2.3897e−08 7.0881e−08 3.2555e−07 2.0152e−08 1.3853e−06 0.00025306 + 3.0199e−11
Rank 11 9 8 7 6 4 5 3 2 1 10
Mean 3.5887e−01 1.1361e−04 9.6510e−05 1.3805e−04 1.5825e−04 1.5610e−04 2.9330e−04 5.6881e−04 1.3562e−03 6.4772e−03 4.6329e−02
STD 1.4421e−01 8.9412e−05 9.2001e−05 9.6139e−05 1.1127e−04 8.6655e−05 1.9597e−04 2.7592e−04 7.1885e−04 2.8905e−03 2.7570e−02
F7
p-value 3.0199e−11 0.34783 + 0.031466 0.004856 0.0026243 2.4913e−06 1.6132e−10 4.5043e−11 3.0199e−11 3.0199e−11
Rank 11 2 1 3 5 4 6 7 8 9 10
Mean 2.5132e+02 2.9219e−185 1.1857e−141 1.1244e−118 3.6048e−100 1.4965e−77 9.2916e−56 2.1466e−35 5.4301e−18 3.2676e−05 2.0739e+01
STD 6.8198e+01 0.0000e+00 2.7646e−141 3.3811e−118 1.3727e−99 6.4522e−77 1.8339e−55 3.4388e−35 1.0245e−17 2.6015e−05 1.0314e+01
F8
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 1.4071e−01 0.0000e+00 6.0776e−287 5.9898e−241 3.8919e−202 2.7715e−160 1.2285e−114 4.8825e−74 8.2506e−40 6.4935e−15 1.5859e−03
STD 8.2755e−02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 1.3814e−159 3.2487e−114 1.8038e−73 3.5120e−39 9.6320e−15 1.5708e−03
F9
p-value 1.2118e−12 + 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 6.8339e−04 2.2338e−109 1.5971e−91 1.6766e−82 6.9381e−75 6.1302e−63 9.7271e−50 3.5221e−37 7.7232e−25 1.9212e−12 4.4188e−05
STD 5.4258e−04 2.5020e−109 5.2647e−91 6.9631e−82 1.0637e−74 1.2266e−62 2.1087e−49 6.0182e−37 1.2239e−24 2.6680e−12 2.4234e−05
F10
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean −6.3155e+03 −5.0340e+03 −6.2256e+03 −6.4204e+03 −6.4148e+03 −6.5818e+03 −6.6016e+03 −6.7509e+03 −6.8964e+03 −7.1658e+03 −7.1404e+03
STD 3.4840e+02 5.6687e+02 3.9280e+02 2.5980e+02 3.1851e+02 2.6239e+02 2.9080e+02 3.6362e+02 2.4347e+02 3.1249e+02 3.2658e+02
F11
p-value 8.1014e−10 4.6159e−10 5.5727e−10 1.0937e−10 3.4742e−10 8.4848e−09 2.6015e−08 1.3367e−05 0.00044592 + 0.75059
Rank 9 11 10 7 8 6 5 4 3 1 2
Mean 1.8594e+02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 2.8270e+01 8.8521e+01 1.3763e+02
STD 1.5546e+01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 2.3270e+01 9.9619e+00 2.1240e+01
F12
p-value 1.2118e−12 + NaN NaN NaN NaN NaN NaN 1.2118e−12 1.2118e−12 1.2118e−12
Rank 5 1 1 1 1 1 1 1 2 3 4

(continued on next page)

Table 3 (continued).
Algorithms
Functions FPA FO-FPAα=0.1 FO-FPAα=0.2 FO-FPAα=0.3 FO-FPAα=0.4 FO-FPAα=0.5 FO-FPAα=0.6 FO-FPAα=0.7 FO-FPAα=0.8 FO-FPAα=0.9 FO-FPAα=1
Mean 7.8934e+00 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 1.4399e−09 5.3325e−03 5.2468e+00
STD 1.3245e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 9.4232e−10 1.9763e−03 1.1251e+00
F13
p-value 1.2118e−12 + NaN NaN NaN NaN NaN NaN 1.2118e−12 1.2118e−12 1.2118e−12
Rank 5 1 1 1 1 1 1 1 2 3 4
Mean 1.7391e+01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 1.1467e−03 2.1995e+00
STD 5.3572e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 1.5699e−03 6.4779e−01
F14
p-value 1.2118e−12 + NaN NaN NaN NaN NaN NaN NaN 1.2118e−12 1.2118e−12
Rank 4 1 1 1 1 1 1 1 1 2 3
Mean 3.2574e+02 8.3277e−01 6.7497e−01 6.0130e−01 5.9321e−01 5.8056e−01 5.4339e−01 5.4759e−01 4.7390e−01 4.6012e−01 4.0201e+00
STD 1.0779e+03 1.7848e−01 1.0989e−01 1.0994e−01 9.7085e−02 1.1260e−01 1.4278e−01 1.0414e−01 8.4003e−02 9.5070e−02 1.6146e+00

D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889
F15
p-value 3.0199e−11 6.121e−10 2.1947e−08 9.5139e−06 6.7362e−06 0.00011058 0.0094683 0.0014423 0.41191 + 3.0199e−11
Rank 11 9 8 7 6 5 3 4 2 1 10
Mean 1.2086e+05 2.9406e+00 2.9131e+00 2.9078e+00 2.9348e+00 2.8665e+00 2.8626e+00 2.6609e+00 2.5221e+00 2.4890e+00 1.4210e+01
STD 1.4077e+05 1.5905e−01 2.0100e−01 2.5213e−01 1.8224e−01 2.9136e−01 2.5606e−01 2.9031e−01 2.1092e−01 2.9227e−01 5.0445e+00
F16
p-value 3.0199e−11 2.8314e−08 1.1567e−07 2.7829e−07 3.0811e−08 1.1937e−06 4.1178e−06 0.027086 0.57929 + 3.0199e−11
Rank 11 9 7 6 8 5 4 3 2 1 10
Mean 4.5656e+01 2.9687e+01 2.9902e+01 2.9804e+01 2.9406e+01 2.8430e+01 2.8011e+01 2.4322e+01 2.4244e+01 2.2318e+01 4.7180e+01
STD 1.5170e+01 8.8757e−01 2.7064e−02 4.3954e−01 1.8603e+00 2.9068e+00 3.4382e+00 3.1719e+00 2.4067e+00 2.8252e+00 1.0721e+01
F17
p-value 4.0772e−11 4.5043e−11 3.0199e−11 3.6897e−11 4.6159e−10 5.5329e−08 2.1959e−07 0.026077 0.0098834 + 3.0199e−11
Rank 10 7 9 8 6 5 4 3 2 1 11
Mean 2.0866e+01 4.4269e−94 2.0573e−72 2.5183e−61 1.8333e−51 3.4806e−40 7.5457e−29 9.7615e−19 7.7126e−10 6.0264e−02 1.5971e+01
STD 2.3996e+00 4.6479e−94 2.1366e−72 3.7556e−61 3.9211e−51 4.6694e−40 1.2985e−28 9.9747e−19 3.6077e−10 5.2917e−02 2.6941e+00
F18
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 2.1948e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 8.9971e−07 5.4077e−01
STD 3.0675e−01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 8.0712e−07 3.3560e−01
F19
p-value 1.2118e−12 + NaN NaN NaN NaN NaN NaN NaN 1.2118e−12 1.2118e−12
Rank 4 1 1 1 1 1 1 1 1 2 3
Mean 1.2083e+02 9.1983e−185 1.1824e−140 1.9153e−118 8.9473e−99 1.3255e−75 1.0740e−54 2.3714e−34 1.8804e−11 1.8631e−03 7.9682e+00
STD 3.0167e+01 0.0000e+00 2.0839e−140 6.8011e−118 3.0830e−98 6.5584e−75 2.1453e−54 4.5537e−34 7.8562e−11 2.9211e−03 4.0467e+00
F20
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean −1.7399e+01 −1.8173e+01 −2.2232e+01 −2.2715e+01 −2.2698e+01 −2.1943e+01 −2.1011e+01 −1.7048e+01 −1.2735e+01 −1.2465e+01 −1.4228e+01
STD 4.3843e+00 9.9889e−01 8.7849e−01 1.0043e+00 1.0450e+00 9.1587e−01 1.3943e+00 3.0049e+00 2.9258e+00 2.6476e+00 4.1683e+00
F21
p-value 1.3594e−07 3.0199e−11 0.13345 + 0.79585 0.0042259 3.0939e−06 3.0199e−11 3.0199e−11 3.0199e−11 1.0702e−09
Rank 7 6 3 1 2 4 5 8 10 11 9
Mean 2.0152e+01 2.9921e+01 2.9706e+01 2.9564e+01 2.9167e+01 2.7638e+01 2.8192e+01 2.6577e+01 2.3802e+01 2.2318e+01 3.0255e+01
STD 5.1967e+00 2.9853e−02 7.2095e−01 1.6369e+00 2.1121e+00 3.5695e+00 3.1043e+00 3.1870e+00 2.6731e+00 3.2605e+00 5.8934e+00
F22
p-value + 5.5727e−10 7.3803e−10 1.5465e−09 4.1825e−09 6.5261e−07 9.8329e−08 3.0939e−06 0.0016798 0.055546 8.352e−08
Rank 1 10 9 8 7 5 6 4 3 2 11
Mean 7.5896e+06 1.4584e−179 4.7526e−136 5.9656e−114 4.6638e−95 2.6907e−72 1.9682e−50 1.1372e−30 4.4361e−13 2.7015e+00 1.2475e+06
STD 2.7360e+06 0.0000e+00 1.9781e−135 1.9260e−113 1.0523e−94 8.3471e−72 5.7326e−50 1.7115e−30 6.7179e−13 2.3090e+00 6.4293e+05
F23
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 0.0000e+00 −1.2460e−103 −3.7604e−85 −4.8900e−80 −2.7151e−96 −3.4056e−87 −6.6321e−102 −7.8966e−89 −1.5533e−296 0.0000e+00 0.0000e+00
STD 0.0000e+00 6.8234e−103 2.0596e−84 2.6783e−79 1.4523e−95 1.8653e−86 2.2429e−101 4.3251e−88 0.0000e+00 0.0000e+00 0.0000e+00
F24
p-value 1.2118e−12 3.5923e−05 0.41191 + 0.37904 0.70617 0.79585 0.40354 1.7203e−12 1.2118e−12 1.2118e−12
Rank 9 7 2 1 5 3 6 4 8 10 11

(continued on next page)

15
 16
Table 3 (continued).
Algorithms
Functions FPA FO-FPAα=0.1 FO-FPAα=0.2 FO-FPAα=0.3 FO-FPAα=0.4 FO-FPAα=0.5 FO-FPAα=0.6 FO-FPAα=0.7 FO-FPAα=0.8 FO-FPAα=0.9 FO-FPAα=1
Mean 5.5541e+00 2.3890e−87 2.4750e−52 1.5069e−39 7.8534e−29 2.4827e−02 8.3289e−02 9.9224e−02 9.9873e−02 2.6093e−01 2.4480e+00
STD 5.8876e−01 7.3206e−87 1.0787e−51 8.2521e−39 3.1482e−28 1.9362e−02 2.7443e−02 3.5561e−03 6.4714e−10 5.3958e−02 5.9443e−01
F25
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 4.9261e−01 0.0000e+00 4.1468e−03 3.5313e−03 4.8572e−03 9.4626e−03 9.7159e−03 9.7159e−03 9.7159e−03 5.4140e−02 4.7857e−01
STD 4.8227e−03 0.0000e+00 2.5169e−03 2.4149e−03 2.8714e−03 1.3873e−03 4.6649e−11 1.4394e−10 2.6835e−10 1.9354e−02 1.5054e−02
F26
p-value 1.2118e−12 + 1.2118e−12 1.2118e−12 4.5736e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12
Rank 11 1 3 2 4 5 6 7 8 9 10
Mean 9.9829e−01 8.8724e+00 6.2693e+00 5.5926e+00 4.6997e+00 2.7346e+00 2.4846e+00 2.1008e+00 1.7389e+00 1.2903e+00 1.0001e+00
STD 1.4466e−03 3.3409e+00 3.5840e+00 2.3424e+00 2.6078e+00 1.6538e+00 8.5903e−01 9.2914e−01 6.9389e−01 4.3801e−01 6.4334e−03
F27
3.0199e−11 6.0658e−11 7.3891e−11 3.0199e−11 8.1527e−11 5.4941e−11 3.1589e−10 4.1997e−10 1.6132e−10 4.084e−05

D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889
p-value +
Rank 1 11 10 9 8 7 6 5 4 3 2
Mean 6.6576e−04 2.2662e−03 1.0985e−03 8.6270e−04 9.4145e−04 6.4147e−04 7.1642e−04 6.9553e−04 7.3883e−04 6.9345e−04 8.6943e−04
STD 1.4943e−04 1.5236e−03 9.0719e−04 3.8603e−04 5.2502e−04 1.2985e−04 1.5812e−04 1.6294e−04 1.1028e−04 1.0160e−04 1.6601e−04
F28
p-value 0.28378 2.3768e−07 0.023243 0.019112 0.0086844 + 0.040595 0.12967 0.000812 0.029205 3.2555e−07
Rank 2 11 10 7 9 1 5 4 6 3 8
Mean −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00
STD 9.1095e−08 4.0460e−05 3.0660e−07 2.6296e−07 3.4338e−07 2.1629e−07 1.6270e−07 1.9521e−07 1.6089e−07 2.7415e−07 1.4484e−08
F29
p-value 4.639e−05 3.0199e−11 1.6132e−10 1.0702e−09 2.0152e−08 5.0922e−08 3.4742e−10 3.8202e−10 7.3803e−10 3.4742e−10 +
Rank 2 11 10 9 7 4 5 6 3 8 1
Mean 3.9789e−01 8.9025e−01 6.3491e−01 5.8678e−01 5.1569e−01 5.0974e−01 4.7860e−01 4.8132e−01 4.4965e−01 4.0551e−01 3.9789e−01
STD 5.8700e−12 4.9426e−01 1.7488e−01 1.8775e−01 1.0149e−01 8.8740e−02 6.1965e−02 1.2139e−01 4.7273e−02 6.2517e−03 3.3354e−09
F30
p-value + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 1 11 10 9 8 7 5 6 4 3 2
Mean 3.0000e+00 1.6683e+01 1.1461e+01 7.6768e+00 7.1590e+00 7.0393e+00 6.8470e+00 5.3935e+00 4.1067e+00 3.1442e+00 3.0000e+00
STD 3.1311e−09 1.0295e+01 8.9843e+00 5.8783e+00 6.8766e+00 7.0768e+00 5.1393e+00 2.8111e+00 1.7342e+00 1.7032e−01 1.3243e−05
F31
p-value + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 1 11 10 9 8 7 6 5 4 3 2
Mean −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01
STD 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16
F32
p-value + NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN
Rank 1 1 1 1 1 1 1 1 1 1 1
Mean −3.2960e+00 −2.5977e+00 −3.0025e+00 −3.0359e+00 −3.0455e+00 −3.0429e+00 −3.0797e+00 −3.1088e+00 −3.0832e+00 −3.1250e+00 −3.1199e+00
STD 1.9661e−02 3.4493e−01 2.2653e−01 1.9206e−01 1.7999e−01 1.6284e−01 1.8555e−01 1.5967e−01 1.9188e−01 1.5259e−01 1.3540e−01
F33
p-value + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 4.0772e−11 8.9934e−11 4.0772e−11 4.0772e−11 5.4941e−11 4.9752e−11
Rank 1 11 10 9 7 8 6 4 5 2 3
Mean −1.0146e+01 −3.9176e+00 −4.7707e+00 −4.7911e+00 −4.9970e+00 −5.1429e+00 −5.5983e+00 −5.7278e+00 −5.5050e+00 −5.3611e+00 −5.4975e+00
STD 8.1113e−03 1.3689e+00 2.1739e+00 1.3244e+00 1.5893e+00 1.8489e+00 2.5138e+00 2.3430e+00 1.9606e+00 2.2905e+00 2.3692e+00
F34
p-value + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 1 11 10 9 8 7 3 2 4 6 5
Mean −1.0309e+01 −3.6129e+00 −4.5858e+00 −5.0464e+00 −5.3065e+00 −6.0693e+00 −5.7287e+00 −5.9188e+00 −5.5365e+00 −5.4039e+00 −5.0963e+00
STD 1.3954e−01 9.5656e−01 1.4411e+00 1.1968e+00 1.7202e+00 2.0314e+00 1.5370e+00 1.7334e+00 1.4360e+00 1.7346e+00 1.4547e+00
F35
p-value + 3.0199e−11 3.6897e−11 3.0199e−11 3.6897e−11 1.2057e−10 3.0199e−11 3.6897e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 1 11 10 9 7 2 4 3 5 6 8
Mean −1.0399e+01 −3.6960e+00 −4.7093e+00 −5.2042e+00 −5.4924e+00 −5.8908e+00 −5.6738e+00 −6.0019e+00 −5.8439e+00 −5.4496e+00 −5.2180e+00
STD 1.5282e−01 1.5285e+00 1.2197e+00 1.7392e+00 1.8563e+00 2.2118e+00 2.1366e+00 1.9782e+00 1.8408e+00 1.2894e+00 1.5131e+00
F36
p-value + 3.0199e−11 3.0199e−11 1.0937e−10 6.0658e−11 6.0658e−11 6.0658e−11 7.7725e−09 8.1014e−10 3.0199e−11 3.4742e−10
Rank 1 11 10 9 6 3 5 2 4 7 8
+/= /− 9/ 1/ 26 17/ 1/ 18 6/ 2/ 28 7/ 0/ 29 5/ 2/ 29 6/ 2/ 28 5/ 2/ 29 5/ 3/ 28 4/ 2/ 30 6/ 1/ 29 2/ 1/ 33
Average rank 7.0278 5.4167 5.1389 4.8333 5 4.3333 4.6667 4.6944 5.0833 5.6111 7.4444
Table 4
Comparison of simulation results of FPA and FO-FPA algorithms with number of terms 12 and α changed from 0.1 to 1 for uni-model and multi-model functions with dimension 30 as well as for fixed-dimensional
multi-modal functions.
Algorithms
Functions FPA FO-FPAα=0.1 FO-FPAα=0.2 FO-FPAα=0.3 FO-FPAα=0.4 FO-FPAα=0.5 FO-FPAα=0.6 FO-FPAα=0.7 FO-FPAα=0.8 FO-FPAα=0.9 FO-FPAα=1
Mean 2.0542e+03 1.2350e−139 6.0813e−118 4.0783e−106 2.2659e−94 3.1660e−75 4.5040e−54 4.0748e−34 4.6120e−17 2.8154e−04 1.4017e+02
STD 7.2876e+02 1.1337e−139 1.2674e−117 1.7156e−105 4.1741e−94 1.2171e−74 1.3720e−53 7.6977e−34 5.9782e−17 2.1656e−04 7.9180e+01
F1
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 2.9404e+01 1.5173e−70 7.5221e−60 2.2414e−54 7.3005e−48 8.2283e−39 7.1189e−28 7.3698e−18 4.2893e−09 1.9233e−02 1.3968e+01
STD 4.9789e+00 5.0712e−71 5.2495e−60 2.6112e−54 9.8468e−48 1.2420e−38 8.8691e−28 5.3719e−18 2.4944e−09 1.4888e−02 4.7600e+00
F2
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 1.4789e+03 3.9655e−139 6.2579e−117 1.3069e−105 4.6473e−93 1.1552e−74 1.9295e−53 2.1104e−33 1.7848e−16 5.2103e−04 1.9301e+02

D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889
STD 4.0380e+02 3.2781e−139 1.9110e−116 4.1229e−105 1.5768e−92 3.3815e−74 3.4193e−53 3.9634e−33 2.1781e−16 3.6536e−04 8.9688e+01
F3
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 2.3949e+01 1.9992e−70 1.6585e−59 7.7249e−54 2.3259e−47 3.5721e−38 2.0988e−27 4.6922e−17 3.0414e−08 4.6689e−02 9.7194e+00
STD 2.8292e+00 1.0365e−70 1.1418e−59 7.0758e−54 3.4871e−47 4.7171e−38 2.5066e−27 4.2711e−17 2.4827e−08 2.4275e−02 2.8677e+00
F4
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 3.3292e+05 2.8934e+01 2.8906e+01 2.8907e+01 2.8890e+01 2.8891e+01 2.8906e+01 2.8894e+01 2.8888e+01 2.8917e+01 5.3024e+03
STD 1.5873e+05 1.6522e−02 3.7445e−02 2.9920e−02 5.2865e−02 5.1123e−02 3.6771e−02 3.7929e−02 3.4393e−02 3.8735e−02 3.4240e+03
F5
p-value 3.0199e−11 5.5329e−08 0.011228 0.023243 0.33285 0.37904 0.016285 0.49178 + 0.0083146 3.0199e−11
Rank 11 9 6 7 2 3 5 4 1 8 10
Mean 1.7825e+03 5.8685e+00 5.3302e+00 5.3116e+00 5.2853e+00 5.3429e+00 5.1573e+00 5.0416e+00 4.8744e+00 4.3796e+00 1.6149e+02
STD 4.6470e+02 4.9062e−01 4.6513e−01 4.2913e−01 3.8642e−01 5.2100e−01 5.5042e−01 5.5160e−01 6.1080e−01 7.0132e−01 8.5273e+01
F6
p-value 3.0199e−11 3.4971e−09 2.154e−06 3.3242e−06 2.6784e−06 2.4913e−06 9.2113e−05 0.00025306 0.0055699 + 3.0199e−11
Rank 11 9 7 6 5 8 4 3 2 1 10
Mean 3.5887e−01 1.0398e−04 1.1703e−04 1.4859e−04 1.5615e−04 2.1751e−04 3.0041e−04 6.9642e−04 1.3712e−03 7.7758e−03 4.4461e−02
STD 1.4421e−01 1.1489e−04 1.1960e−04 9.9428e−05 1.1023e−04 1.7765e−04 1.7923e−04 3.5576e−04 6.8757e−04 3.6982e−03 2.3698e−02
F7
p-value 3.0199e−11 + 0.37108 0.0094683 0.02236 0.0035012 1.729e−06 1.3289e−10 6.0658e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 2.5132e+02 1.5228e−140 6.5596e−119 2.2048e−107 1.0655e−94 1.6182e−76 7.1332e−56 7.5900e−35 6.1567e−18 3.0318e−05 2.0131e+01
STD 6.8198e+01 1.1337e−140 1.3443e−118 8.6647e−107 3.0458e−94 6.7878e−76 1.3285e−55 1.8561e−34 5.4512e−18 1.9478e−05 1.0461e+01
F8
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 1.4071e−01 7.8551e−286 1.8712e−241 3.0761e−219 9.2377e−195 2.8021e−158 1.3670e−114 5.5171e−74 2.6321e−39 1.1785e−14 1.1116e−03
STD 8.2755e−02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 8.3875e−158 3.7783e−114 1.3431e−73 5.5271e−39 1.8222e−14 1.3727e−03
F9
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 6.8339e−04 1.0830e−79 1.4273e−73 9.6193e−72 5.3616e−69 1.8829e−61 1.6460e−49 3.9936e−37 7.6875e−25 1.6052e−12 6.0945e−05
STD 5.4258e−04 8.6579e−80 3.6415e−73 1.1984e−71 7.0683e−69 3.8902e−61 4.7794e−49 7.8394e−37 9.1333e−25 1.5106e−12 5.1064e−05
F10
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean −6.3155e+03 −5.2107e+03 −6.1703e+03 −6.4363e+03 −6.5021e+03 −6.5655e+03 −6.5947e+03 −6.7356e+03 −6.8005e+03 −7.0286e+03 −7.0770e+03
STD 3.4840e+02 6.3103e+02 2.7848e+02 2.8014e+02 2.2898e+02 2.8530e+02 2.5185e+02 2.9763e+02 2.9573e+02 2.6301e+02 2.3118e+02
F11
p-value 2.2273e−09 5.5727e−10 5.4941e−11 7.3803e−10 1.5465e−09 2.6015e−08 1.0105e−08 1.2493e−05 0.00017836 0.34783 +
Rank 9 11 10 8 7 6 5 4 3 2 1
Mean 1.8594e+02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 4.0051e+01 9.0678e+01 1.3598e+02
STD 1.5546e+01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 2.3825e+01 9.8306e+00 1.5466e+01
F12
p-value 1.2118e−12 + NaN NaN NaN NaN NaN NaN 1.2118e−12 1.2118e−12 1.2118e−12
Rank 5 1 1 1 1 1 1 1 2 3 4

(continued on next page)

17
 18
Table 4 (continued).
Algorithms
Functions FPA FO-FPAα=0.1 FO-FPAα=0.2 FO-FPAα=0.3 FO-FPAα=0.4 FO-FPAα=0.5 FO-FPAα=0.6 FO-FPAα=0.7 FO-FPAα=0.8 FO-FPAα=0.9 FO-FPAα=1
Mean 7.8934e+00 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 1.7168e−09 4.7316e−03 5.3237e+00
STD 1.3245e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 1.0288e−09 1.8978e−03 1.4745e+00
F13
p-value 1.2118e−12 + NaN NaN NaN NaN NaN NaN 1.2118e−12 1.2118e−12 1.2118e−12
Rank 5 1 1 1 1 1 1 1 2 3 4
Mean 1.7391e+01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 5.9212e−17 9.6881e−04 2.2976e+00
STD 5.3572e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 2.4172e−16 9.8962e−04 7.1347e−01
F14
p-value 1.2118e−12 + NaN NaN NaN NaN NaN NaN 0.1608 1.2118e−12 1.2118e−12
Rank 5 1 1 1 1 1 1 1 2 3 4
Mean 3.2574e+02 7.3887e−01 5.7916e−01 5.6190e−01 5.4877e−01 5.4902e−01 5.4146e−01 4.6159e−01 4.4976e−01 4.6815e−01 4.3185e+00
STD 1.0779e+03 1.4130e−01 1.3219e−01 1.2201e−01 1.1648e−01 1.0271e−01 1.1297e−01 1.1464e−01 1.0058e−01 1.1079e−01 1.5468e+00

D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889
F15
p-value 3.0199e−11 5.5727e−10 0.0005264 0.00042175 0.0021566 0.00049818 0.0027548 0.79585 + 0.94696 3.0199e−11
Rank 11 9 8 7 5 6 4 2 1 3 10
Mean 1.2086e+05 2.9689e+00 2.9881e+00 2.9615e+00 2.9141e+00 2.7940e+00 2.8227e+00 2.5719e+00 2.5586e+00 2.3141e+00 1.2878e+01
STD 1.4077e+05 9.1823e−02 2.4862e−03 1.4971e−01 2.1099e−01 3.0948e−01 3.1450e−01 3.0689e−01 3.3262e−01 3.2163e−01 4.4927e+00
F16
p-value 3.0199e−11 7.3891e−11 3.0199e−11 1.9568e−10 2.4386e−09 2.3168e−06 4.444e−07 0.0076171 0.004856 + 3.0199e−11
Rank 11 8 9 7 6 4 5 3 2 1 10
Mean 4.5656e+01 2.9721e+01 2.9253e+01 2.9564e+01 2.9257e+01 2.7373e+01 2.7661e+01 2.4561e+01 2.4438e+01 2.2863e+01 4.4750e+01
STD 1.5170e+01 1.0073e+00 1.6202e+00 1.2649e+00 2.0922e+00 3.8370e+00 3.2997e+00 3.6447e+00 3.0321e+00 3.3279e+00 8.9461e+00
F17
p-value 5.4941e−11 1.411e−09 6.5183e−09 2.4386e−09 1.4294e−08 5.265e−05 9.5139e−06 0.085 0.045146 + 5.4941e−11
Rank 11 9 6 8 7 4 5 3 2 1 10
Mean 2.0866e+01 1.2880e−71 1.0471e−60 3.3316e−55 7.0163e−49 9.4562e−40 6.7829e−29 1.0012e−18 8.2800e−10 6.6621e−02 1.4880e+01
STD 2.3996e+00 5.0829e−72 8.1140e−61 3.7071e−55 7.8365e−49 1.1223e−39 7.1407e−29 8.6492e−19 5.1853e−10 9.8817e−02 3.7685e+00
F18
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 2.1948e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 8.6599e−07 5.8277e−01
STD 3.0675e−01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 6.2949e−07 3.3726e−01
F19
p-value 1.2118e−12 + NaN NaN NaN NaN NaN NaN NaN 1.2118e−12 1.2118e−12
Rank 4 1 1 1 1 1 1 1 1 2 3
Mean 1.2083e+02 7.4065e−140 6.4811e−118 1.0147e−106 4.5502e−94 4.4947e−76 6.7505e−53 1.1351e−28 3.9394e−11 7.6705e−04 7.2920e+00
STD 3.0167e+01 7.8371e−140 1.2561e−117 2.1203e−106 1.4805e−93 1.4950e−75 3.1594e−52 6.2169e−28 1.7315e−10 9.1136e−04 4.4083e+00
F20
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean −1.7399e+01 −1.8392e+01 −2.2289e+01 −2.2552e+01 −2.2362e+01 −2.1851e+01 −2.0794e+01 −1.6225e+01 −1.3001e+01 −1.1258e+01 −1.3762e+01
STD 4.3843e+00 9.8558e−01 8.7119e−01 8.0800e−01 7.7839e−01 1.1906e+00 1.2967e+00 3.5158e+00 3.3477e+00 3.4614e+00 4.0477e+00
F21
p-value 1.7294e−07 3.0199e−11 0.14128 + 0.49178 0.015638 7.0881e−08 4.0772e−11 3.3384e−11 3.0199e−11 3.8202e−10
Rank 7 6 3 1 2 4 5 8 10 11 9
Mean 2.0152e+01 2.9548e+01 2.9604e+01 2.9440e+01 2.9647e+01 2.7929e+01 2.8040e+01 2.5007e+01 2.3819e+01 2.1311e+01 3.0633e+01
STD 5.1967e+00 1.7592e+00 1.0380e+00 1.7669e+00 1.0532e+00 3.4388e+00 3.2564e+00 3.3703e+00 2.8700e+00 3.0086e+00 5.0899e+00
F22
p-value + 1.6947e−09 1.1737e−09 2.0338e−09 9.7555e−10 3.2555e−07 1.7294e−07 0.00021265 0.0015969 0.29727 1.0105e−08
Rank 1 8 9 7 10 5 6 4 3 2 11
Mean 7.5896e+06 2.1932e−135 1.0237e−113 3.3348e−102 1.9884e−89 1.0809e−70 1.0832e−49 3.4801e−30 8.3788e−13 2.0665e+00 1.1679e+06
STD 2.7360e+06 1.7741e−135 1.1497e−113 1.1482e−101 4.7412e−89 4.8646e−70 3.9855e−49 4.9932e−30 1.4230e−12 1.4791e+00 7.9845e+05
F23
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 0.0000e+00 −5.5686e−92 −4.1549e−97 −2.0234e−87 −2.4025e−95 −7.8428e−86 −8.0763e−81 −1.9143e−89 −6.8448e−162 0.0000e+00 0.0000e+00
STD 0.0000e+00 3.0500e−91 2.2757e−96 1.1076e−86 1.3159e−94 4.2957e−85 3.2191e−80 1.0485e−88 3.7459e−161 0.0000e+00 0.0000e+00
F24
p-value 1.2118e−12 0.00018916 0.0044272 0.69522 0.12967 0.05012 + 0.66273 2.3657e−12 1.2118e−12 1.2118e−12
Rank 9 5 7 3 6 2 1 4 8 10 11

(continued on next page)

Table 4 (continued).
Algorithms
Functions FPA FO-FPAα=0.1 FO-FPAα=0.2 FO-FPAα=0.3 FO-FPAα=0.4 FO-FPAα=0.5 FO-FPAα=0.6 FO-FPAα=0.7 FO-FPAα=0.8 FO-FPAα=0.9 FO-FPAα=1
Mean 5.5541e+00 1.1978e−65 2.0243e−46 1.9587e−32 4.2621e−23 2.6014e−02 9.0878e−02 9.7471e−02 9.9873e−02 2.7965e−01 2.4371e+00
STD 5.8876e−01 5.5980e−65 6.6810e−46 1.0375e−31 2.3296e−22 2.6831e−02 2.0339e−02 1.0842e−02 8.2123e−10 6.5669e−02 6.0042e−01
F25
p-value 3.0199e−11 + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 11 1 2 3 4 5 6 7 8 9 10
Mean 4.9261e−01 0.0000e+00 5.1055e−03 4.7446e−03 4.4803e−03 9.1219e−03 9.7159e−03 9.7159e−03 9.7159e−03 5.2129e−02 4.6634e−01
STD 4.8227e−03 0.0000e+00 3.2300e−03 2.6552e−03 2.4196e−03 1.5635e−03 8.5846e−11 4.2584e−11 2.0819e−10 2.1075e−02 4.5909e−02
F26
p-value 1.2118e−12 + 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12 1.2118e−12
Rank 11 1 4 3 2 5 7 6 8 9 10
Mean 9.9829e−01 8.0178e+00 5.9733e+00 4.5291e+00 3.3484e+00 2.7752e+00 2.1988e+00 1.8644e+00 1.7946e+00 1.4286e+00 1.0322e+00
STD 1.4466e−03 3.5193e+00 3.3810e+00 2.9183e+00 2.3555e+00 1.5923e+00 1.1805e+00 9.9355e−01 8.3790e−01 4.3275e−01 1.8130e−01
F27
3.0199e−11 3.0199e−11 5.4941e−11 4.9752e−11 3.0199e−11 4.0772e−11 1.9568e−10 4.1997e−10 6.6955e−11 7.7387e−06

D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889
p-value +
Rank 1 11 10 9 8 7 6 5 4 3 2
Mean 6.6576e−04 2.2320e−03 9.6047e−04 7.7194e−04 7.9763e−04 8.2614e−04 7.4363e−04 7.2204e−04 7.2847e−04 7.2398e−04 8.5366e−04
STD 1.4943e−04 1.6355e−03 5.9755e−04 2.1176e−04 2.7362e−04 2.7172e−04 2.5384e−04 1.3299e−04 1.4903e−04 1.1469e−04 1.4375e−04
F28
p-value + 4.3106e−08 0.11199 0.09049 0.13345 0.011711 0.36322 0.21702 0.18577 0.21702 2.959e−05
Rank 1 11 10 6 7 8 5 2 4 3 9
Mean −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00 −1.0316e+00
STD 9.1095e−08 1.5897e−05 2.9341e−07 1.7565e−07 5.3783e−07 2.1424e−07 1.7954e−07 2.3395e−07 3.1471e−07 5.2951e−07 8.4679e−09
F29
p-value 3.5708e−06 3.0199e−11 1.7769e−10 6.0658e−11 6.722e−10 1.85e−08 4.9752e−11 6.6955e−11 3.6897e−11 1.7769e−10 +
Rank 2 11 8 4 7 3 5 6 9 10 1
Mean 3.9789e−01 1.0091e+00 5.7501e−01 5.8544e−01 5.4596e−01 5.2662e−01 4.6659e−01 4.6928e−01 4.4525e−01 4.0458e−01 3.9789e−01
STD 5.8700e−12 9.4695e−01 1.5850e−01 1.9633e−01 1.1996e−01 2.2199e−01 5.5337e−02 6.0725e−02 4.0914e−02 6.0921e−03 7.4226e−09
F30
p-value + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 1 11 9 10 8 7 5 6 4 3 2
Mean 3.0000e+00 1.2668e+01 1.2672e+01 8.3382e+00 6.2500e+00 5.6825e+00 6.4368e+00 5.4438e+00 3.6975e+00 3.1594e+00 3.0000e+00
STD 3.1311e−09 8.3191e+00 8.4964e+00 6.9652e+00 4.2669e+00 3.2788e+00 4.3002e+00 5.1304e+00 8.9508e−01 2.0226e−01 1.2321e−05
F31
p-value + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 1 10 11 9 7 6 8 5 4 3 2
Mean −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01 −3.0048e−01
STD 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16 2.2584e−16
F32
p-value + NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN
Rank 1 1 1 1 1 1 1 1 1 1 1
Mean −3.2960e+00 −2.3572e+00 −2.8912e+00 −3.0455e+00 −3.0657e+00 −3.0390e+00 −3.0611e+00 −3.0887e+00 −3.0434e+00 −3.0652e+00 −3.1188e+00
STD 1.9661e−02 3.9715e−01 2.6468e−01 1.6305e−01 1.4580e−01 1.7755e−01 1.6096e−01 1.5200e−01 2.1069e−01 1.9211e−01 1.0559e−01
F33
p-value + 3.0199e−11 3.0199e−11 3.0199e−11 3.3384e−11 3.0199e−11 3.3384e−11 3.6897e−11 3.3384e−11 3.0199e−11 3.3384e−11
Rank 1 11 10 7 4 9 6 3 8 5 2
Mean −1.0146e+01 −3.3405e+00 −4.1129e+00 −4.9713e+00 −5.0332e+00 −5.0316e+00 −5.0945e+00 −4.9712e+00 −5.3509e+00 −5.1340e+00 −4.6000e+00
STD 8.1113e−03 1.7069e+00 1.3140e+00 2.2461e+00 2.0683e+00 2.3344e+00 2.1362e+00 2.1802e+00 1.9342e+00 2.0534e+00 1.5994e+00
F34
p-value + 3.0199e−11 3.0199e−11 3.0199e−11 7.3891e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11
Rank 1 11 10 7 5 6 4 8 2 3 9
Mean −1.0309e+01 −3.5317e+00 −4.5167e+00 −4.8075e+00 −5.0700e+00 −5.4819e+00 −5.0662e+00 −5.9602e+00 −5.4032e+00 −5.6217e+00 −5.1941e+00
STD 1.3954e−01 1.0268e+00 1.3464e+00 1.3366e+00 1.3103e+00 1.3696e+00 1.5917e+00 1.7717e+00 1.8377e+00 2.1573e+00 1.6556e+00
F35
p-value + 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.0199e−11 3.6897e−11 1.6132e−10 1.6132e−10 6.0658e−11 3.0199e−11
Rank 1 11 10 9 7 4 8 2 5 3 6
Mean −1.0399e+01 −3.7756e+00 −5.2720e+00 −5.3817e+00 −6.0518e+00 −5.9012e+00 −6.2229e+00 −6.3877e+00 −6.0316e+00 −5.9907e+00 −5.3536e+00
STD 1.5282e−01 8.7045e−01 1.6740e+00 1.4771e+00 1.9341e+00 1.5412e+00 2.1879e+00 2.2836e+00 1.5369e+00 1.8400e+00 1.3318e+00
F36
p-value + 3.0199e−11 3.0199e−11 3.3384e−11 2.1544e−10 3.0199e−11 9.9186e−11 2.6015e−08 3.3384e−11 3.0199e−11 3.0199e−11
Rank 1 11 10 8 4 7 3 2 5 6 9
+/= /− 10/ 0/ 26 18/ 0/ 18 5/ 3/ 28 6/ 2/ 28 5/ 4/ 27 5/ 2/ 29 6/ 1/ 29 5/ 5/ 26 4/ 2/ 30 4/ 4/ 28 3/ 0/ 33
Average rank 7.0556 5.2778 5.1667 4.6389 4.5 4.6944 4.8333 4.6944 5.25 5.75 7.5

19
 20
Table 5
Comparison of simulation results of FPA and FO-FPA algorithms for 100 and 500 dimensional benchmark functions.
Algorithms
Functions Benchmarks of dimension 100 Benchmarks of dimension 500
Functions FPA FO-FPA4,0.4 FO-FPA4,0.5 FO-FPA8,0.5 FO-FPA8,0.6 FO-FPA12,0.4 FPA FO-FPA4,0.4 FO-FPA4,0.5 FO-FPA8,0.5 FO-FPA8,0.6 FO-FPA12,0.4
Best 8.7877e+03 6.4386e−102 3.6678e−76 1.0143e−80 1.8361e−56 6.4211e−97 6.5586e+04 1.8545e−107 1.5149e−79 1.9189e−79 1.2738e−55 4.0264e−78
Mean 1.3337e+04 4.5321e−103 3.8735e−77 4.1741e−76 1.4809e−53 2.6278e−93 7.8823e+04 8.0610e−100 9.4608e−76 2.8869e−74 2.9997e−52 9.0354e−74
F1
Worst 1.7733e+04 1.4416e−107 8.3587e−82 3.6951e−75 2.2791e−52 3.5122e−92 1.1222e+05 2.2259e−98 1.0260e−74 4.2710e−73 5.1477e−51 9.9032e−73
STD 2.2023e+03 1.2355e−102 9.3461e−77 8.5992e−76 4.1247e−53 6.9425e−93 1.1698e+04 4.0647e−99 2.6275e−75 8.4133e−74 9.8423e−52 2.1131e−73
Best 7.9241e+01 4.4901e−50 5.1361e−38 5.8432e−40 8.6984e−29 1.8520e−48 3.7389e+02 2.9278e−53 1.5480e−40 2.8080e−39 1.2822e−27 9.9454e−39
Mean 1.0063e+02 2.1648e−51 4.8051e−39 1.4538e−38 2.4059e−27 3.0596e−47 4.9183e+02 4.4799e−51 1.3107e−38 7.0078e−38 2.3602e−26 3.1712e−37
F2
Worst 1.1710e+02 3.9598e−54 7.4718e−42 1.4765e−37 9.7369e−27 3.3545e−46 5.6676e+02 4.6726e−50 1.1468e−37 3.2035e−37 1.7533e−25 1.7336e−36

D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889
STD 1.0911e+01 8.1467e−51 9.6550e−39 2.7649e−38 2.6194e−27 6.2103e−47 3.8855e+01 8.9335e−51 2.2527e−38 8.6188e−38 3.7287e−26 4.9075e−37
Best 1.3793e+04 1.7316e−99 2.0265e−73 1.0514e−78 4.3402e−56 5.9181e−95 2.3115e+05 1.0386e−102 1.7590e−77 1.1327e−76 2.3505e−53 1.3392e−74
Mean 2.0346e+04 1.3035e−100 7.9557e−75 1.3728e−74 3.4895e−52 1.4074e−91 4.4376e+05 1.2436e−98 5.3482e−74 1.5298e−73 6.4354e−50 3.2238e−70
F3
Worst 2.8820e+04 3.4889e−105 2.9575e−79 1.6259e−73 5.0569e−51 1.6743e−90 7.7047e+05 1.7624e−97 1.2553e−72 8.1858e−73 1.2583e−48 9.3668e−69
STD 3.8980e+03 3.4727e−100 3.7022e−74 3.2845e−74 9.9385e−52 4.0775e−91 1.2569e+05 3.5054e−98 2.2858e−73 2.2343e−73 2.2776e−49 1.7085e−69
Best 2.6406e+01 2.4079e−51 3.2919e−38 1.6251e−40 3.6814e−28 1.0772e−48 3.6251e+01 1.4356e−53 2.9720e−40 2.8889e−40 1.8210e−28 1.3489e−38
Mean 3.4981e+01 5.9090e−52 3.0303e−39 1.8527e−38 4.8942e−27 2.6136e−47 4.4988e+01 8.9952e−51 2.5640e−38 5.5483e−38 9.9816e−27 3.5166e−37
F4
Worst 4.4296e+01 5.0936e−54 6.1581e−41 1.1626e−37 2.8045e−26 9.5163e−47 5.2325e+01 1.7756e−49 3.8495e−37 3.7975e−37 5.5897e−26 3.9282e−36
STD 4.0610e+00 6.5470e−52 6.2691e−39 2.9705e−38 6.0281e−27 2.7854e−47 4.0777e+00 3.3303e−50 6.9941e−38 8.6627e−38 1.2460e−26 7.7145e−37
Best 2.4241e+06 9.8947e+01 9.8929e+01 9.8854e+01 9.8808e+01 9.8760e+01 1.6786e+07 4.9880e+02 4.9875e+02 4.9873e+02 4.9873e+02 4.9874e+02
Mean 4.1071e+06 9.8894e+01 9.8879e+01 9.8890e+01 9.8896e+01 9.8878e+01 3.2648e+07 4.9887e+02 4.9886e+02 4.9886e+02 4.9886e+02 4.9885e+02
F5
Worst 5.9281e+06 9.8787e+01 9.8817e+01 9.8940e+01 9.8954e+01 9.8939e+01 5.7444e+07 4.9894e+02 4.9893e+02 4.9891e+02 4.9891e+02 4.9892e+02
STD 9.1273e+05 3.9241e−02 3.3442e−02 2.3094e−02 3.5131e−02 3.5599e−02 9.6875e+06 3.3207e−02 4.1680e−02 3.8210e−02 3.9470e−02 4.1216e−02
Best 8.7083e+03 2.3119e+01 2.3449e+01 1.9942e+01 2.0123e+01 1.9256e+01 6.2070e+04 1.1959e+02 1.1912e+02 1.1579e+02 1.1835e+02 1.1915e+02
Mean 1.3058e+04 2.2066e+01 2.2064e+01 2.1805e+01 2.2169e+01 2.1915e+01 7.8881e+04 1.2150e+02 1.2131e+02 1.2107e+02 1.2122e+02 1.2100e+02
F6
Worst 1.7856e+04 2.0616e+01 2.0401e+01 2.3251e+01 2.3250e+01 2.3516e+01 9.7278e+04 1.2345e+02 1.2286e+02 1.2283e+02 1.2255e+02 1.2221e+02
STD 2.2066e+03 6.5473e−01 6.9832e−01 8.1721e−01 7.3542e−01 9.8855e−01 9.7843e+03 9.7113e−01 9.1234e−01 1.4674e+00 1.0815e+00 7.1479e−01
Best 2.8215e+00 4.0580e−04 8.9621e−04 1.5084e−05 2.1715e−05 1.7162e−06 1.9753e+02 3.6839e−05 2.4956e−05 1.7234e−05 4.4835e−05 1.6987e−05
Mean 6.0609e+00 1.7188e−04 2.0178e−04 2.3780e−04 3.1316e−04 1.7762e−04 2.8491e+02 1.7791e−04 2.2182e−04 1.9426e−04 3.8888e−04 2.1302e−04
F7
Worst 1.0862e+01 6.4862e−06 2.3679e−05 5.7206e−04 9.2814e−04 5.9014e−04 4.4114e+02 4.6245e−04 8.0048e−04 6.4855e−04 1.5179e−03 6.7030e−04
STD 2.0024e+00 1.1849e−04 1.7586e−04 1.5962e−04 2.0773e−04 1.3075e−04 6.7296e+01 1.2672e−04 1.6397e−04 1.5137e−04 2.9692e−04 1.7771e−04
Best 3.7240e+03 6.0557e−101 1.0326e−75 2.9638e−81 8.2726e−57 4.3873e−97 1.4324e+05 3.5267e−105 6.2604e−79 4.3485e−78 2.7277e−54 1.4636e−78
Mean 5.7792e+03 3.2325e−102 4.2565e−77 6.6697e−76 1.5310e−53 1.4698e−93 1.9497e+05 1.3264e−99 5.1090e−76 3.3836e−74 2.7306e−52 2.6360e−72
F8
Worst 7.4418e+03 4.2225e−107 5.1018e−83 1.1894e−74 1.6295e−52 1.7503e−92 2.4215e+05 3.2494e−98 6.6581e−75 2.9742e−73 2.0065e−51 7.7594e−71
STD 8.3175e+02 1.1345e−101 1.8831e−76 2.3751e−75 3.4243e−53 3.7759e−93 2.7741e+04 5.9516e−99 1.2711e−75 7.8159e−74 4.7349e−52 1.4158e−71
Best 2.9434e+00 6.1547e−207 6.9940e−159 1.8457e−167 2.6630e−119 7.0791e−200 1.4850e+02 1.4731e−217 1.6337e−166 5.2140e−165 7.0004e−118 3.8593e−162
Mean 6.5683e+00 3.7615e−208 4.6026e−160 2.8141e−156 2.4475e−112 3.5184e−190 2.6208e+02 2.1802e−208 9.4921e−159 5.4705e−154 1.3661e−110 7.1877e−153
F9
Worst 1.5877e+01 3.3676e−220 1.8838e−172 6.5612e−155 3.4203e−111 1.0550e−188 6.6147e+02 4.5379e−207 2.4634e−157 1.6066e−152 2.0402e−109 1.4164e−151
STD 2.6031e+00 0.0000e+00 1.6796e−159 1.2288e−155 7.4348e−112 0.0000e+00 9.7767e+01 0.0000e+00 4.5022e−158 2.9313e−153 4.2709e−110 2.7202e−152
Best 9.3499e−05 5.2994e−78 1.2009e−62 6.1868e−66 9.5762e−52 3.0505e−71 5.7140e−04 1.0029e−83 6.4550e−68 8.8220e−67 2.1183e−53 3.8574e−64
Mean 2.0195e−03 3.2099e−79 9.9519e−64 1.2967e−62 1.2663e−49 2.9452e−68 2.8803e−03 1.1773e−78 3.1424e−64 4.3415e−63 6.8081e−50 1.3651e−60
F10
Worst 1.1325e−02 1.7788e−82 4.5040e−67 1.0935e−61 8.8495e−49 7.1612e−67 1.1147e−02 1.5664e−77 6.7168e−63 2.8210e−62 1.4362e−48 2.4724e−59
STD 2.0221e−03 9.6940e−79 2.5304e−63 2.4302e−62 2.3431e−49 1.3007e−67 2.3198e−03 3.1359e−78 1.2181e−63 6.8465e−63 2.5985e−49 4.6786e−60
Best −1.3978e+04 −1.2993e+04 −1.3341e+04 −1.4641e+04 −1.5027e+04 −1.5594e+04 −3.3150e+04 −3.4373e+04 −3.5296e+04 −3.6543e+04 −3.5727e+04 −3.4988e+04
Mean −1.2781e+04 −1.3861e+04 −1.3956e+04 −1.3882e+04 −1.4119e+04 −1.3872e+04 −2.9819e+04 −3.2650e+04 −3.3223e+04 −3.3156e+04 −3.3763e+04 −3.3097e+04
F11
Worst −1.1696e+04 −1.4793e+04 −1.6001e+04 −1.2799e+04 −1.3165e+04 −1.2605e+04 −2.7538e+04 −2.9004e+04 −3.0773e+04 −3.1360e+04 −3.2306e+04 −3.0782e+04
STD 5.0449e+02 3.8438e+02 4.8418e+02 4.2095e+02 4.3124e+02 6.3113e+02 1.3503e+03 1.3202e+03 9.7855e+02 1.2707e+03 1.0134e+03 1.1339e+03

(continued on next page)

Table 5 (continued).
Algorithms
Functions Benchmarks of dimension 100 Benchmarks of dimension 500
Functions FPA FO-FPA4,0.4 FO-FPA4,0.5 FO-FPA8,0.5 FO-FPA8,0.6 FO-FPA12,0.4 FPA FO-FPA4,0.4 FO-FPA4,0.5 FO-FPA8,0.5 FO-FPA8,0.6 FO-FPA12,0.4
Best 7.8190e+02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 4.8126e+03 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
Mean 8.5558e+02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 4.9368e+03 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
F12
Worst 9.5428e+02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 5.0965e+03 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
STD 4.1312e+01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 6.0973e+01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
Best 6.4162e+00 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 7.1632e+00 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16
Mean 8.0427e+00 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.7341e+00 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16
F13
Worst 9.9533e+00 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 1.1416e+01 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16
STD 8.8775e−01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 9.9482e−01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
Best 7.4112e+01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 5.1928e+02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00

D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889
Mean 1.0504e+02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 6.9293e+02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
F14
Worst 1.4086e+02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 8.3270e+02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
STD 1.7268e+01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 8.4553e+01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
Best 1.6511e+03 1.0627e+00 1.0897e+00 7.6629e−01 7.0329e−01 7.5745e−01 6.8670e+05 1.0619e+00 1.0656e+00 1.0007e+00 1.0651e+00 1.0639e+00
Mean 1.1086e+05 9.1634e−01 9.3357e−01 9.1805e−01 9.0469e−01 9.2389e−01 4.0909e+06 1.1122e+00 1.1103e+00 1.1086e+00 1.1038e+00 1.1116e+00
F15
Worst 5.4427e+05 7.5998e−01 8.2079e−01 1.0207e+00 1.0490e+00 1.0634e+00 1.6953e+07 1.1425e+00 1.1380e+00 1.1523e+00 1.1398e+00 1.1535e+00
STD 1.1339e+05 7.8137e−02 7.0967e−02 6.3886e−02 7.8852e−02 7.8162e−02 3.6220e+06 2.3230e−02 1.8683e−02 2.8636e−02 1.9450e−02 2.1493e−02
Best 8.1411e+05 9.9936e+00 9.9941e+00 9.9284e+00 9.9519e+00 9.9009e+00 1.5428e+07 4.9936e+01 4.9934e+01 4.9935e+01 4.9941e+01 4.9904e+01
Mean 3.2759e+06 9.9858e+00 9.9855e+00 9.9842e+00 9.9848e+00 9.9801e+00 4.0178e+07 4.9980e+01 4.9982e+01 4.9981e+01 4.9981e+01 4.9982e+01
F16
Worst 1.2762e+07 9.9618e+00 9.9368e+00 9.9908e+00 9.9920e+00 9.9910e+00 7.0431e+07 4.9991e+01 4.9991e+01 4.9993e+01 4.9991e+01 4.9990e+01
STD 2.6508e+06 5.6969e−03 1.0099e−02 1.2426e−02 7.6862e−03 2.1225e−02 1.6353e+07 1.3825e−02 1.1971e−02 1.2482e−02 1.0421e−02 1.4886e−02
Best 2.0109e+02 9.9937e+01 9.9934e+01 9.9030e+01 9.9756e+01 9.9312e+01 1.5287e+03 4.9897e+02 4.9849e+02 4.9905e+02 4.9944e+02 4.9975e+02
Mean 3.0190e+02 9.9851e+01 9.9856e+01 9.9819e+01 9.9859e+01 9.9837e+01 1.9427e+03 4.9979e+02 4.9974e+02 4.9973e+02 4.9979e+02 4.9983e+02
F17
Worst 4.0548e+02 9.9072e+01 9.9365e+01 9.9919e+01 9.9912e+01 9.9926e+01 2.3917e+03 4.9991e+02 4.9990e+02 4.9990e+02 4.9988e+02 4.9990e+02
STD 4.7910e+01 1.5181e−01 9.7089e−02 1.6778e−01 3.8151e−02 1.0847e−01 2.0714e+02 1.9244e−01 2.8811e−01 2.2261e−01 1.0427e−01 3.9740e−02
Best 6.5276e+01 4.2036e−52 4.3385e−39 5.6348e−41 2.0604e−29 8.0224e−50 3.3472e+02 3.7612e−54 1.1809e−41 7.8648e−40 1.5859e−28 3.4792e−40
Mean 8.0604e+01 5.5639e−53 4.9915e−40 1.2480e−39 3.1135e−28 1.3173e−48 3.8558e+02 2.7720e−52 2.4987e−39 4.6360e−39 1.6242e−27 2.1702e−38
F18
Worst 9.1783e+01 1.9610e−55 8.8770e−42 7.1031e−39 3.8742e−27 5.2707e−48 4.4690e+02 4.5870e−51 1.4680e−38 2.3411e−38 7.3694e−27 2.0557e−37
STD 6.2158e+00 1.1293e−52 8.2415e−40 1.6290e−39 7.1282e−28 1.5334e−48 2.6439e+01 8.2788e−52 3.8013e−39 4.7194e−39 1.8901e−27 3.7906e−38
Best 7.6808e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 4.4319e+01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
Mean 9.6596e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 5.0989e+01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
F19
Worst 1.0955e+01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 5.6683e+01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
STD 7.4625e−01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 2.9801e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
Best 9.8688e+02 6.5331e−97 1.5784e−70 6.2003e−76 8.6218e−52 6.5562e−92 5.1444e+18 6.2648e−80 9.7735e−56 1.0713e−52 4.5887e−31 9.3541e−53
Mean 1.6277e+03 5.1600e−98 5.9873e−72 9.2713e−72 3.5957e−46 1.2253e−88 2.1057e+19 4.9735e−75 1.7164e−48 7.3262e−49 1.8274e−26 3.4995e−47
F20
Worst 2.7767e+03 3.5466e−102 1.2027e−77 1.7969e−70 6.6182e−45 1.5022e−87 4.6714e+19 9.3982e−74 4.5584e−47 9.2585e−48 5.0646e−25 9.3708e−46
STD 3.8754e+02 1.3296e−97 2.8791e−71 3.3422e−71 1.4042e−45 3.5214e−88 1.1422e+19 1.7429e−74 8.3047e−48 1.9383e−48 9.2249e−26 1.7077e−46
Best −6.9796e+01 −3.9493e+01 −4.1005e+01 −4.8356e+01 −4.7275e+01 −4.8319e+01 −4.1225e+02 −1.0549e+02 −1.0210e+02 −1.0073e+02 −4.3347e+02 −1.0747e+02
Mean −4.9771e+01 −4.4449e+01 −4.4493e+01 −4.4147e+01 −4.3073e+01 −4.4360e+01 −2.2332e+02 −9.4474e+01 −9.3619e+01 −9.0830e+01 −1.0470e+02 −9.3297e+01
F21
Worst −2.5979e+01 −4.9238e+01 −4.9095e+01 −4.0230e+01 −3.8687e+01 −3.9314e+01 −9.0652e+01 −8.6203e+01 −8.5464e+01 −8.0853e+01 −8.3275e+01 −7.6632e+01
STD 1.1117e+01 2.1454e+00 1.8810e+00 2.3642e+00 2.2756e+00 2.0270e+00 7.0867e+01 4.7414e+00 4.2020e+00 5.1021e+00 6.2378e+01 6.3446e+00
Best 1.2259e+02 9.9934e+01 9.9927e+01 9.9188e+01 9.8959e+01 9.8986e+01 8.9496e+02 4.9899e+02 4.9922e+02 4.9919e+02 4.9941e+02 4.9898e+02
Mean 1.5009e+02 9.9822e+01 9.9844e+01 9.9829e+01 9.9795e+01 9.9785e+01 9.9734e+02 4.9981e+02 4.9974e+02 4.9980e+02 4.9980e+02 4.9975e+02
F22
Worst 1.8329e+02 9.8955e+01 9.9511e+01 9.9903e+01 9.9924e+01 9.9917e+01 1.0956e+03 4.9992e+02 4.9991e+02 4.9992e+02 4.9988e+02 4.9989e+02
STD 1.4892e+01 1.9010e−01 7.6760e−02 1.2497e−01 2.2051e−01 2.3289e−01 4.8410e+01 1.5971e−01 2.0564e−01 1.3916e−01 8.6481e−02 2.2669e−01
Best 1.4513e+08 4.4200e−96 1.4924e−71 3.6419e−76 3.8955e−52 2.0085e−91 3.1111e+09 1.4130e−101 5.3598e−77 4.0162e−74 6.9333e−51 1.2865e−73
Mean 2.8842e+08 1.9641e−97 8.5341e−73 6.9682e−72 1.2185e−48 1.4634e−88 4.4114e+09 1.2286e−96 4.5895e−71 3.8474e−70 2.8986e−47 5.1973e−69
F23
Worst 4.5237e+08 8.2299e−104 1.4765e−76 8.2469e−71 2.0408e−47 1.8453e−87 6.5685e+09 2.7922e−95 8.2594e−70 9.0713e−69 7.8581e−46 4.9934e−68
STD 7.2257e+07 8.0761e−97 2.9633e−72 1.6902e−71 4.0646e−48 4.1699e−88 9.0416e+08 5.0853e−96 1.5781e−70 1.6460e−69 1.4301e−46 1.2409e−68

(continued on next page)

21
 22
D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889
Table 5 (continued).
Algorithms
Functions Benchmarks of dimension 100 Benchmarks of dimension 500
Functions FPA FO-FPA4,0.4 FO-FPA4,0.5 FO-FPA8,0.5 FO-FPA8,0.6 FO-FPA12,0.4 FPA FO-FPA4,0.4 FO-FPA4,0.5 FO-FPA8,0.5 FO-FPA8,0.6 FO-FPA12,0.4
Best 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
Mean 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
F24
Worst 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
STD 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
Best 1.0519e+01 2.0991e−28 9.9873e−02 1.7208e−07 1.4084e−02 3.8718e−39 2.6098e+01 3.6338e−43 1.4280e−06 1.0532e−03 2.9690e−02 5.1020e−03
Mean 1.2344e+01 7.2428e−30 2.7804e−02 2.4893e−02 7.9338e−02 6.3486e−27 3.0496e+01 1.0280e−26 2.0867e−02 2.4045e−02 8.9407e−02 3.1114e−02
F25
Worst 1.4215e+01 1.3926e−41 8.4666e−08 9.9873e−02 9.9873e−02 1.2083e−25 3.5715e+01 3.0252e−25 9.9873e−02 9.9873e−02 9.9873e−02 9.9873e−02
STD 9.0186e−01 3.8298e−29 2.9370e−02 2.6390e−02 3.0257e−02 2.4814e−26 2.4114e+00 5.5206e−26 1.9815e−02 2.3880e−02 2.0052e−02 2.2667e−02
Best 4.9885e−01 9.7159e−03 9.7159e−03 2.2644e−03 9.7159e−03 0.0000e+00 4.9997e−01 8.7529e−04 7.3115e−03 8.8856e−03 9.7159e−03 5.8823e−03
Mean 4.9959e−01 5.3280e−03 9.7132e−03 9.4087e−03 9.7159e−03 5.2887e−03 4.9999e−01 6.1856e−03 9.5255e−03 9.6882e−03 9.7159e−03 9.4965e−03
F26
Worst 4.9984e−01 7.7099e−04 9.6354e−03 9.7159e−03 9.7159e−03 9.7159e−03 5.0000e−01 9.7159e−03 9.7159e−03 9.7159e−03 9.7159e−03 9.7159e−03
STD 2.2409e−04 3.0044e−03 1.4704e−05 1.3873e−03 4.2323e−11 3.1311e−03 5.9076e−06 3.4072e−03 5.4988e−04 1.5159e−04 1.5453e−11 8.4697e−04
 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889 23

multi-modal ones. The details of the considered functions are F16, F17, F21, F22, and F24. The FO-FPA variants with different
listed in Table 1. values of alpha across all the considered number of terms (r = 4,
A series of analysis is conducted among the FO-FPA and the 8, 12) expose a better performance than FPA while FO-FPA with
basic FPA as well as state-of-the-art methods in the literature. For alpha equaled 0.1 is the best variant for functions of F18, F19, F20,
fulfilling a fair comparison among the considered techniques, the F23, F25, and F26 and followed by FO-FPA with alpha of values 0.2
number of the populations and maximum iterations are set to 30 to 0.6. FPA shows a better performance for a fixed dimensional
and 500, respectively. The studied algorithms are implemented benchmark function set (F27-F36), whereas the FO-FPA is the
for 30 independent times; they are initiated with the same ran- more robust and reliable for the larger-scale ones (F1–F27).
dom solutions to have a fair judgment. A Wilcoxon rank-sum Based on the average values of the overall ranks across the re-
test with significant difference 0.05 is performed to substantiate garded functions, FO-FPA variants have realized the best average
the significant difference of the depicted results [43]. The rank-
rank values especially with derivative orders of 0.4 and 0.5 in the
based test is also performed to study the proposed algorithm
case of utilizing 4 terms from past events as clear in Table 2. For
quality. Moreover the mean, and standard deviation (STD) values
considering 8 historical terms as shown in Table 3, the derivative
are computed across the independent runs.
orders of 0.5 and 0.6 are the most effective values while the
• Mean: It is preformed to find the center of fitness values and derivative order of 0.4 is the successful one with accounting 12
it is formulated in Eq. (16). terms from memory as reported in Table 4. Accordingly, FO-
25 FPA with derivative order values of 0.4 to 0.6 has a noticeable
1 ∑ performance with a significant difference in comparison with the
Mean = Fitbi (16)
25 basic FPA. The overall results confirm the strength of merging the
i=1
FO calculus with the FPA in modifying its quality and consistency.
• Standard deviation (STD): It is applied to compute the dis-
persion of the fitness values from Mean and it is defined in
4.1.2. Scalability test
Eq. (17).
 In this subsection, the FPA and the FO-FPA with the recom-
 25
1 ∑ mended derivative orders are examined with scalable twenty-six
STD = √ (Fitbi − Mean)2 , (17) benchmarks set with dimensions of 100, and 500 to test the
24 quality of the introduced algorithm with increasing the problem
i=1
size. The best, worst, mean and STD values of the results are per-
4.1.1. Comparison among FPA and FO-FPA formed for the analysis target. The obtained results are reported
In this part, the comparison among FPA and the novel ap- in Table 5 in the Appendix, The results reveal the superiority
proach of FO-FPA is carried out based on 36 benchmarks with of FO-FPA variants with increasing the dimension of the studied
dimensions of 30. The uni-model benchmarks set (F1–F10) is tasks. The best, worst, mean and STD values by FO-FPA variants
utilized to test the exploitation tendencies of the algorithms demonstrate their outperforming for the FPA in achieving the
whereas set of multi-modal functions (F11–F36) are employed for closest values for the optimal ones of the considered functions
examining the exploration propensity and the balance among the
with highest consistency. FPA exposes the best performance for
exploration and exploitation capabilities. The mean, standard de-
only one function (F21) out of twenty-six ones and shows the
viation (STD) and Wilcoxon rank-sum with significant difference
same response for F24. The results confirmed on that the FO
0.05 as well as the average ranking values across the regarded
calculus has a positive influence on the FPA behavior especially
functions are conducted for the analysis stage as in Tables 2–4, for
with the multi-dimensional optimization problem.
all the considered number of terms r of 4, 8 and 12, respectively.
The best results appeared with bold font.
By inspecting the results of the 36 benchmarks with the di- 4.1.3. Comparison with state-of-the-art algorithms
mension of 30 in Tables 2–4, it can be seen that FO-FPA variants In this part, the performance of FO-FPA is evaluated in com-
outperform the FPA in several functions based on the mean, STD parable with well-regarded state-of-the-art techniques such as
values, and the overall ranks as well as FO-FPA variants show cuckoo search (CS) [8], grey wolf Optimizer (GWO) [7], and salp
a significant difference versus FPA in several tasks based on the swarm algorithm (SSA) [9]. The values of the parameters of each
obtained p − v alues. method are similar to the original implementation of each of
For the uni-model benchmark suite (F1–F10) in Tables 2–4, method. Moreover, the number of iterations and population size
FPA is ranked as 11 with a significant difference from the best are tuned as 500, and 30 for all algorithms. The obtained mean,
based on Wilcoxon rank-sum where its p − v alues > 0.05 (big and STD values by the early mentioned algorithms for twenty-six
evidence to reject the null null hypothesis). Meanwhile FO-FPA
functions of dimension 1000 are reported in Table 6.
variants occupy the first ranks for this set of the functions. This
From the reported mean and STD values by the listed algo-
observation proves that considering some terms from memory
rithms, it is noticed that the fractional order flower pollination
based on the FO calculus feature has a favorable impact in the
variants occupy the first places in achieving the most accurate
exploitation inclination of the classical FPA, by this way the local
and consistent results in comparable with FPA, SSA, CS and GWO
search ability is improved. It is obvious that the FO-FPA variants
with derivative order α from 0.1 to 0.6 have the first places for not only for most of the uni-modal functions (F1–F10) but also
this set of the functions, whatever the considered number of for most of the multi-model functions suite. GWO exhibits a best
terms (r = 4 or 8 or 12). performance for F6, F17 and F22 while SSA is the best for F11
As per the results of the multi-model benchmarks set (F11– and FPA has the first place for F21 meanwhile FO-FPA variants
F36) in Tables 2–4, the FO-FPA variants especially with alpha have a capability to provide the closest results for the optimal
of range 0.1 to 0.6 have attained the first ranking places in solutions in twenty-one functions. FO-FPA with considering four
comparable to the FPA that has an average rank of 6 across this terms of memory and derivative order of 0.4 can be ordered as a
benchmarks suite. The FO-FPA with derivative orders 0.1 to 0.6 first among these variants and then followed by FO-FPA with r is
and the number of historical terms are 4 or 8 or 12 have the same 12 and α equaled 0.4 in the second place then FO-FPA with r = 4
response for functions of F12, F13, F14, and F 19. For tasks of F15, and 8 in cases of derivative orders are 0.5 and 0.6.
24 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889

Table 6
Comparison of simulation results of FPA, FO-FPA and state of-the-art algorithms for 1000 dimensional benchmark functions.
Algorithms
Functions FPA FO-FPA4,0.4 FO-FPA4,0.5 FO-FPA8,0.5 FO-FPA8,0.6 FO-FPA12,0.4 SSA CS GWO
Mean 1.6824e+05 2.1660e−100 4.0582e−76 2.5441e−73 7.5067e−52 3.1346e−92 2.3660e+05 1.6625e+05 2.4271e−01
F1 STD 2.5284e+04 6.8316e−100 6.3004e−76 1.0209e−72 1.8507e−51 5.6361e−92 1.2642e+04 1.4934e+04 5.4400e−02
Mean 8.0957e+02 9.3874e−51 6.3584e−38 1.0775e−37 3.8908e−26 6.6936e−46 1.1884e+03 1.0000e+10 6.7987e−01
F2 STD 1.0253e+02 1.7234e−50 8.6359e−38 2.1043e−37 4.9014e−26 1.1481e−45 3.1567e+01 0.0000e+00 3.8935e−01
Mean 1.9853e+06 3.1574e−98 5.8349e−72 2.1781e−70 2.3931e−48 2.8682e−89 6.2616e+06 8.8786e+06 1.4482e+06
F3 STD 5.1082e+05 1.3425e−97 2.7418e−71 1.1414e−69 1.2714e−47 5.1912e−89 2.6285e+06 1.1300e+06 2.4194e+05
Mean 5.1563e+01 1.3804e−50 3.3252e−38 1.4029e−37 2.7618e−26 1.0390e−46 4.5012e+01 4.3167e+01 7.7580e+01
F4 STD 4.1079e+00 4.0614e−50 5.2439e−38 3.0124e−37 4.1224e−26 8.9490e−47 3.1980e+00 3.0228e+00 3.9985e+00
Mean 7.7706e+07 9.9884e+02 9.9885e+02 9.9885e+02 9.9884e+02 9.9884e+02 1.2217e+08 6.7586e+07 1.0583e+03
F5 STD 1.8834e+07 4.6705e−02 3.6727e−02 4.1593e−02 3.3944e−02 3.9671e−02 1.0909e+07 1.2572e+07 3.3621e+01
Mean 1.5798e+05 2.4611e+02 2.4591e+02 2.4583e+02 2.4566e+02 2.4582e+02 2.3496e+05 1.6437e+05 2.0285e+02
F6 STD 1.7402e+04 1.1722e+00 9.7153e−01 1.0897e+00 9.7888e−01 1.2966e+00 1.2858e+04 1.4222e+04 3.1571e+00
Mean 1.1304e+03 1.8510e−04 2.3811e−04 3.8903e−04 3.6408e−04 2.2093e−04 1.7819e+03 1.0306e+03 1.5184e−01
F7 STD 2.8035e+02 1.4683e−04 2.1983e−04 2.7752e−04 2.0009e−04 1.4011e−04 2.3079e+02 2.0540e+02 3.0978e−02
Mean 8.0215e+05 4.8569e−97 3.0074e−75 1.4706e−74 1.6314e−51 3.4169e−91 1.1242e+06 7.6495e+05 9.8778e−01
F8 STD 1.1405e+05 2.6593e−96 8.1627e−75 3.9048e−74 3.8119e−51 6.1056e−91 5.2717e+04 5.7729e+04 2.7206e−01
Mean 1.1405e+03 3.9853e−206 1.7296e−155 6.4108e−153 3.3681e−109 1.9959e−189 1.7567e+03 1.0438e+03 7.0816e−06
F9 STD 2.9067e+02 0.0000e+00 8.5725e−155 3.4705e−152 1.2107e−108 0.0000e+00 1.8222e+02 2.0935e+02 4.2468e−06
Mean 4.2799e−03 9.4363e−79 7.8030e−64 1.7196e−62 1.3437e−49 2.1104e−68 1.7886e−03 7.3292e−06 3.2432e−02
F10 STD 3.0462e−03 1.9864e−78 2.5707e−63 5.0015e−62 2.6339e−49 7.1538e−68 1.5852e−03 1.6426e−05 6.8096e−02
Mean −4.2370e+04 −4.6354e+04 −4.7342e+04 −4.7181e+04 −4.8267e+04 −4.6952e+04 −8.8074e+04 −7.0011e+04 −8.6660e+04
F11 STD 1.7453e+03 1.3988e+03 2.0575e+03 1.5410e+03 1.8729e+03 1.7928e+03 5.5885e+03 1.4354e+03 1.3840e+04
Mean 1.0158e+04 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 7.6633e+03 9.2413e+03 1.9774e+02
F12 STD 1.1309e+02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 1.8531e+02 1.4283e+02 4.1566e+01
Mean 8.3936e+00 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 8.8818e−16 1.4521e+01 1.4926e+01 1.8119e−02
F13 STD 7.1884e−01 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 1.8027e−01 1.1530e+00 2.5693e−03
Mean 1.4968e+03 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 2.1205e+03 1.4997e+03 5.7437e−02
F14 STD 2.4238e+02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 9.8371e+01 1.0735e+02 8.1122e−02
Mean 8.9618e+06 1.1436e+00 1.1408e+00 1.1414e+00 1.1417e+00 1.1394e+00 1.0461e+07 1.0000e+10 1.2173e+00
F15 STD 4.9326e+06 1.0016e−02 1.5034e−02 1.3047e−02 1.0971e−02 1.0821e−02 2.7031e+06 0.0000e+00 2.3958e−01
Mean 8.1127e+07 9.9974e+01 9.9977e+01 9.9976e+01 9.9979e+01 9.9980e+01 1.4966e+08 1.0000e+10 1.1927e+02
F16 STD 3.2785e+07 2.0754e−02 1.9571e−02 1.7115e−02 1.3250e−02 1.8470e−02 1.6268e+07 0.0000e+00 6.4025e+00
Mean 3.9471e+03 9.9978e+02 9.9979e+02 9.9976e+02 9.9969e+02 9.9976e+02 4.9519e+03 4.1477e+03 9.5614e+02
F17 STD 3.3529e+02 1.7197e−01 1.0521e−01 1.7861e−01 2.5092e−01 1.4925e−01 1.9421e+02 2.6016e+02 7.3983e+00
Mean 7.7518e+02 7.5343e−52 2.9812e−39 2.2302e−38 2.4109e−27 3.8130e−47 7.3233e+02 7.7481e+02 1.2110e+00
F18 STD 4.7589e+01 9.4814e−52 4.4910e−39 4.8372e−38 2.3853e−27 5.2595e−47 2.6799e+01 3.2015e+01 1.6053e+00
Mean 1.0489e+02 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 1.0239e+02 1.0080e+02 1.8118e−04
F19 STD 5.2057e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 2.3462e+00 3.4693e+00 4.9756e−05
Mean 1.9302e+22 2.4509e−59 2.8477e−36 2.7272e−35 3.6359e−10 1.0513e−52 1.9790e+04 1.0000e+10 7.9773e+03
F20 STD 5.6389e+21 1.3125e−58 7.5394e−36 1.0017e−34 1.9906e−09 5.3046e−52 5.3291e+02 0.0000e+00 7.5228e+02
Mean −4.6698e+02 −1.1973e+02 −1.2207e+02 −1.2323e+02 −1.2399e+02 −1.2204e+02 3.5598e+02 −4.5225e+01 1.1536e+02
F21 STD 9.9465e+01 6.3318e+00 7.6884e+00 7.8841e+00 8.2843e+00 7.9021e+00 7.5831e+00 7.1689e+01 5.6430e+01
Mean 2.0642e+03 9.9982e+02 9.9980e+02 9.9975e+02 9.9972e+02 9.9974e+02 2.1779e+03 1.9520e+03 9.4557e+02
F22 STD 1.0172e+02 5.9360e−02 1.0617e−01 2.1560e−01 2.1477e−01 1.7583e−01 9.0672e+01 9.7257e+01 7.7052e+00
Mean 9.8555e+09 1.2074e−95 1.6424e−71 2.2182e−70 1.0987e−47 1.1908e−86 7.0514e+09 1.0000e+10 6.5051e+02
F23 STD 1.3877e+09 3.9301e−95 2.6893e−71 5.7096e−70 2.0859e−47 6.2187e−86 9.0207e+08 0.0000e+00 1.6425e+02
Mean 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
F24 STD 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00 0.0000e+00
Mean 4.2797e+01 5.0769e−25 3.7521e−02 2.3142e−02 8.9399e−02 1.4851e−20 5.3463e+01 4.5304e+01 1.9489e+00
F25 STD 4.0100e+00 1.9476e−24 3.1941e−02 2.4576e−02 2.4515e−02 8.1208e−20 1.7803e+00 1.7457e+00 1.6715e−01
Mean 5.0000e−01 6.0946e−03 9.5328e−03 9.6808e−03 9.7159e−03 5.7708e−03 4.9999e−01 5.0000e−01 4.7403e−01
F26 STD 1.6070e−06 2.8213e−03 8.8469e−04 1.9207e−04 4.6426e−11 2.8140e−03 8.4546e−07 5.5581e−07 6.8360e−03

4.1.4. Acceleration convergence trend analysis for the optimal values for all the regarded functions except F11.
In this section, the impact of merging the FO calculus memory According for the depicted curves, the FO calculus enhance not
feature with FPA in improving the acceleration convergence trend only the exploitation capability of the FPA but also it improves
is discussed. Therefore, the mean convergence curves by FO-FPA the harmonization stage among the exploration and exploitation
with recommended derivative orders versus FPA and CS, GWO, tendencies.
and SSA are exhibited in Figs. 2 and 3. By inspecting the con-
vergence curves of Figs. 2 and 3, we can observe the stagnation 4.2. Experimental series 2: Image segmentation
performance of FPA nearly for all the functions, while GWO has
the best acceleration trend for F11 then followed by CS in the To investigate the efficiency of the FO-FPA approach, we ap-
second place. The FO-FPA variants exhibit a fastest decaying rate plied it as a multilevel threshold (MLT) image segmentation
 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889 25

Table 7
PSNR values at different α and r.
0.4 0.5 0.6
4 8 12 4 8 12 4 8 12
Im1 12.220 12.761 12.482 12.845 12.364 12.289 12.737 13.352 13.031
Im2 15.149 15.279 15.012 14.661 15.143 14.770 15.281 14.993 14.551
Im3 14.278 14.485 14.055 14.428 14.380 14.469 14.330 14.921 14.392
Im4 15.012 14.736 14.869 14.858 14.873 14.833 14.919 15.088 14.907
Im5 13.902 14.208 14.282 14.145 14.228 14.076 14.268 14.168 14.315
2
Im6 15.154 14.855 14.774 15.221 15.328 14.936 15.322 15.042 15.316
Im7 11.513 11.467 11.327 11.443 11.435 11.627 11.372 11.506 11.596
Im8 14.370 14.428 14.390 14.238 14.385 14.277 14.506 14.431 14.500
Im9 12.415 11.875 12.309 12.298 11.981 12.011 11.887 12.524 12.076
Im10 13.773 13.732 13.883 13.862 13.847 13.992 13.357 15.173 13.722
Im1 21.535 21.666 21.591 21.396 21.809 21.505 21.515 21.972 21.601
Im2 22.119 22.119 22.372 22.223 21.864 22.159 22.005 21.981 22.034
Im3 20.811 19.630 20.662 20.564 21.392 20.131 20.643 21.750 20.609
Im4 21.627 21.708 21.783 21.421 21.527 21.798 21.789 21.582 21.506
Im5 22.687 22.280 22.693 21.676 23.175 22.329 22.114 23.203 22.396
12
Im6 21.375 21.277 20.538 20.687 20.754 20.960 21.252 20.983 20.649
Im7 21.690 21.007 21.258 21.515 21.335 21.002 21.485 21.545 21.285
Im8 18.953 19.464 19.997 19.270 18.961 19.493 19.381 19.379 19.398
Im9 21.146 21.350 21.048 21.138 21.245 21.096 20.788 21.966 21.664
Im10 20.144 19.835 20.443 20.207 20.926 19.961 20.228 19.947 19.997
Im1 29.523 29.595 29.923 29.800 29.343 29.843 29.701 29.950 29.890
Im2 29.082 28.867 28.753 28.444 28.429 28.840 28.992 28.701 28.634
Im3 29.453 29.103 28.922 28.826 29.415 29.372 29.542 29.857 28.965
Im4 29.613 29.156 29.464 29.405 29.414 29.470 29.304 29.485 29.634
Im5 28.857 28.169 28.916 28.524 29.235 28.432 29.031 29.973 28.761
28
Im6 29.213 28.794 29.078 29.132 29.406 29.356 29.212 29.371 28.897
Im7 29.545 29.762 29.658 29.687 29.333 29.703 29.431 29.643 29.492
Im8 28.402 28.393 28.793 28.853 28.805 29.212 29.218 28.931 29.009
Im9 29.140 28.896 29.477 29.349 28.788 29.606 28.940 30.102 29.179
Im10 27.98 28.8249 30.096 28.0301 28.5318 28.7125 28.3601 28.7836 29.2252

method as a real-world application [44]. Since the image segmen- to determine a set of K thresholds (tk ) that divide I into its classes
tation is considered a prepossessing step in different applications (i.e., Ck , k = 1, 2, . . . , K ) and this defined in Eq. (18):
for example, medical image [45], remote sensing [46], infrared C0 = {Iij | 0 ≤ Iij ≤ t1 − 1},
images [47,48] and others [49–51,51,52]. The main aim of the
C1 = {Iij | t1 ≤ Iij ≤ t2 − 1},
MLT image segmentation is to allocate the threshold values that (18)
divide the images into number of groups according to some ...
criteria [53]. CK = {Iij | tK ≤ Iij ≤ L − 1}
Recently, several MH methods have been developed and ap-
where Iij is the gray level of I and L represents the maximum gray
plied to MLT image segmentation problem. For example, in [54],
value of I.
authors provided the moth flame optimization (MFO) and whale
The MLT problem can be considered as optimization problem
optimization algorithm (WOA) as a two MLT image segmentation by maximizing the objective function (Fit) as defined in Eq. (19):
methods and both of them used the Otsu as fitness function.
Resma et al. [55] introduced a MIT method based on the krill herd
optimization (KHO) in which the solutions are assessed using the t1∗ , t2∗ , . . . , tK∗ = arg max Fit(t1 , . . . , tK ) (19)
t1 ,...,tK
Otsu and Kapur functions. In [56], authors presented an image
segmentation method based on the chaotic bat algorithm (CBA). In this study, the fuzzy entropy is used as Fit and it is defined in
They also compared the CBA approach with other MH methods Eq. (20)
and it showed better performance. Resma et al. [55] developed K
∑
an MIT approach based on the krill herd optimization (KHO). In Fit(t1 , . . . , tK ) = Hi (20)
this approach, the solutions were evaluated by using the Otsu and k=1
Kapur as fitness functions. Whereas, the differential evolution and
L−1 L−1
Tsallis Fuzzy entropy is proposed an image segmentation method ∑ pi × µk (i) pi × µk (i) ∑
in [57]. Hk = − × ln( ), Pk = pi × µk (i) (21)
Pk Pk
However, these methods still need more improvements since i=0 i=0

most of them has its own limitations that can effect on the quality ⎧
of the final solution to MLT problem. Therefore, we will apply the ⎨1 l ≤ a1
l−c1
proposed FO-FPA method to avoid these limitations. In addition, µ1 (l) = a1 −c1
a1 ≤ l ≤ c1 ,
0 l > c1
⎩
we used the fuzzy entropy as fitness function since it has ability
to remove the greyness ambiguities existing in the image [58]. ⎧ (22)
This will lead to improve the quality of the segmented image.
⎨1 l ≤ aK −1
aK −1 < l ≤ cK −1
l−aK
µK (l) =
⎩ cK −aK
4.2.1. Problem definition of multi-level thresholding 0 l > cK −1
In this section, the mathematical formulation for the MLT In Eq. (22), the fuzzy parameters are given by a1 , c1 , . . . ., ak−1 ,
image segmentation is discussed. Consider, the gray-scale image ck−1 and 0 ≤ a1 ≤ c1 ≤ · · · ≤ aK −1 ≤ cK −1 . The threshold
tk = k 2 k , k = 1, 2, . . . , K − 1.
a +c
I which contains K + 1 classes. Therefore, the MLT method aims
26 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889

Table 8
PSNR and SSIM results for each algorithm.
PSNR SSIM
CS GWO SSA FPA FO-FPA CS GWO SSA FPA FO-FPA
Im1 14.254 14.123 16.015 14.813 17.428 0.5235 0.5103 0.6159 0.5472 0.6691
Im2 15.881 15.612 17.464 16.730 18.564 0.4040 0.4023 0.5158 0.4532 0.5395
Im3 12.881 12.605 13.695 13.004 16.889 0.6162 0.6072 0.6678 0.6382 0.7533
Im4 16.211 16.331 15.977 16.820 16.747 0.5448 0.5513 0.5450 0.5768 0.5872
Im5 11.666 11.903 13.959 13.044 15.621 0.2994 0.3153 0.4490 0.3851 0.5430
6
Im6 11.924 12.183 14.593 13.978 17.393 0.3415 0.3617 0.5015 0.4659 0.6054
Im7 11.983 11.822 14.405 12.895 16.592 0.4191 0.4188 0.5790 0.4652 0.6212
Im8 14.489 14.019 16.094 14.606 17.144 0.5921 0.5727 0.6729 0.6008 0.7119
Im9 10.151 10.599 13.179 12.044 16.481 0.5780 0.7024 0.7045 0.6831 0.7755
Im10 14.212 14.424 16.715 15.845 18.688 0.6603 0.6614 0.7621 0.7265 0.8416
Im1 18.151 17.706 18.703 18.412 19.047 0.7146 0.7044 0.7243 0.7164 0.7309
Im2 16.894 16.539 19.318 18.541 20.084 0.4540 0.4567 0.5781 0.5186 0.5892
Im3 15.720 15.964 17.822 17.690 19.094 0.7611 0.7528 0.7867 0.7795 0.8004
Im4 17.698 17.064 18.292 18.476 19.510 0.6005 0.5889 0.6356 0.6419 0.6814
Im5 16.013 16.157 16.787 15.916 18.408 0.5522 0.5653 0.5948 0.5456 0.6740
8
Im6 15.184 15.585 17.558 16.197 19.477 0.5114 0.5342 0.6263 0.5570 0.6865
Im7 15.996 15.544 17.195 16.887 18.793 0.5850 0.5685 0.6387 0.6263 0.6819
Im8 15.153 16.899 17.486 15.842 17.917 0.6480 0.7071 0.7231 0.6757 0.7343
Im9 15.504 15.424 17.592 17.392 18.520 0.8056 0.8046 0.8258 0.8293 0.8351
Im10 19.108 19.316 19.482 18.377 20.271 0.7771 0.7632 0.8460 0.8033 0.8587
Im1 23.013 21.509 23.226 23.371 24.288 0.8378 0.8128 0.8408 0.8463 0.8666
Im2 22.437 22.187 23.264 23.705 23.876 0.6742 0.7036 0.7024 0.7149 0.7355
Im3 21.528 19.667 24.011 23.557 24.298 0.8541 0.8498 0.8678 0.8687 0.8530
Im4 21.667 21.685 23.118 23.183 22.844 0.7395 0.7485 0.7798 0.7824 0.7824
Im5 21.165 21.295 22.384 22.114 22.935 0.7627 0.7845 0.8069 0.7901 0.8174
15
Im6 21.151 20.510 23.817 23.233 24.399 0.7408 0.7220 0.8139 0.7982 0.8185
Im7 21.324 20.229 21.885 22.632 22.889 0.7676 0.7929 0.7653 0.7873 0.8025
Im8 21.823 21.299 22.665 22.958 23.061 0.8248 0.8354 0.8471 0.8509 0.8575
Im9 20.969 18.096 22.038 22.883 23.290 0.8489 0.8542 0.8625 0.8666 0.8836
Im10 21.459 21.467 22.629 23.604 23.652 0.8620 0.8462 0.8744 0.8910 0.9011
Im1 24.529 23.075 24.659 25.443 23.880 0.8716 0.8434 0.8729 0.8860 0.8606
Im2 24.146 24.048 25.094 25.567 24.756 0.7345 0.7562 0.7641 0.7705 0.7408
Im3 23.327 20.658 24.597 25.472 24.783 0.8687 0.8666 0.8781 0.8792 0.8675
Im4 22.894 22.487 24.100 24.475 24.374 0.7736 0.7722 0.8069 0.8146 0.8185
Im5 22.685 22.868 23.854 24.219 23.818 0.8101 0.8275 0.8424 0.8532 0.8400
17
Im6 22.213 22.155 25.307 24.236 25.353 0.7773 0.7747 0.8361 0.8178 0.8364
Im7 22.614 21.414 24.061 24.130 24.321 0.7876 0.8193 0.8181 0.8171 0.8361
Im8 22.681 22.887 23.967 24.133 24.737 0.8414 0.8635 0.8723 0.8704 0.8804
Im9 22.704 19.356 23.235 23.809 24.882 0.8592 0.8549 0.8803 0.8751 0.8982
Im10 23.155 21.930 24.077 23.784 25.935 0.8850 0.8550 0.8892 0.8948 0.9103
Im1 25.236 24.251 26.028 26.420 25.660 0.8830 0.8653 0.8971 0.9032 0.8927
Im2 25.350 24.971 26.311 26.161 25.991 0.7644 0.7881 0.7851 0.7795 0.7718
Im3 24.743 21.786 25.628 25.849 26.799 0.8762 0.8741 0.8864 0.8909 0.8810
Im4 23.709 23.913 24.292 24.617 25.538 0.7928 0.8019 0.8149 0.8201 0.8407
Im5 24.154 23.857 24.747 25.188 25.026 0.8509 0.8525 0.8650 0.8743 0.8722
19
Im6 24.851 23.623 26.559 25.405 25.762 0.8361 0.8093 0.8653 0.8495 0.8510
Im7 24.532 22.666 25.380 25.677 25.500 0.8339 0.8400 0.8461 0.8482 0.8441
Im8 24.152 23.879 24.907 25.596 24.745 0.8766 0.8809 0.8841 0.8906 0.8823
Im9 22.523 20.864 24.831 24.680 25.143 0.8703 0.8711 0.8839 0.8882 0.8887
Im10 24.317 22.756 24.853 25.360 25.080 0.9050 0.8870 0.9057 0.9100 0.8970
Im1 27.401 26.732 28.484 29.199 28.721 0.9151 0.9058 0.9291 0.9386 0.9314
Im2 28.227 28.058 28.039 28.854 27.987 0.8372 0.8647 0.8386 0.8553 0.8456
Im3 26.803 23.930 28.512 29.181 28.482 0.9041 0.9014 0.9031 0.9129 0.9016
Im4 26.752 26.257 27.608 27.534 27.393 0.8606 0.8544 0.8774 0.8768 0.8751
Im5 27.395 26.906 27.558 27.384 27.417 0.9126 0.9111 0.9152 0.9108 0.9127
25
Im6 26.745 27.180 27.900 27.724 28.223 0.8747 0.8864 0.8868 0.8871 0.8912
Im7 27.406 25.971 27.368 28.378 27.902 0.8798 0.8777 0.8815 0.8950 0.8902
Im8 27.203 26.709 27.592 27.991 27.419 0.9118 0.9145 0.9152 0.9212 0.9196
Im9 26.565 24.435 26.701 27.660 27.427 0.9035 0.8934 0.9052 0.9082 0.9109
Im10 27.664 25.956 28.740 29.098 28.317 0.9242 0.9223 0.9288 0.9288 0.9239
Average 21.038 20.389 22.273 22.200 22.993 0.749 0.752 0.787 0.777 0.806

4.2.2. Dataset description 4.2.3. Performance measures

In this experiment, a set of ten natural gray-scale images is In this subsection, a set of measures is used to evaluate the
quality of the segmented image based on the obtained thresh-
used. The original images and their histograms are given in Fig. 4.
old value from each algorithm. These measures include the fit-
This figure depict each image and its histogram which indicates ness value of each algorithm, the Peak-to-noise-ratio (PSNR), and
the amount of information in each image. Structural Similarity Index (SSIM) which defined as:
 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889 27

Table 9
Fitness value and CPU time(s) for each algorithm.
Fitness Time
CS GWO SSA FPA FO-FPA CS GWO SSA FPA FO-FPA
Im1 17.516 17.525 17.664 17.377 17.375 0.5769 0.4637 1.2575 1.5237 1.0537
Im2 17.292 17.289 17.085 17.163 16.815 0.5830 0.4646 1.2781 1.5293 1.0573
Im3 17.087 17.082 17.127 16.995 16.504 0.5520 0.4413 1.2531 1.4590 0.9994
Im4 17.552 17.570 17.344 17.403 16.991 0.5442 0.4377 1.2255 1.4435 0.9965
Im5 15.598 15.593 15.340 15.406 14.945 0.5487 0.4484 1.2456 1.4477 0.9989
6
Im6 15.070 15.081 14.688 14.885 14.012 0.5610 0.4517 1.2526 1.4951 1.0318
Im7 17.621 17.624 17.841 17.514 16.977 0.5492 0.4414 1.2423 1.4550 1.0019
Im8 17.574 16.972 17.389 17.478 17.590 0.4562 0.3472 1.1384 1.1724 0.8155
Im9 17.477 16.834 17.227 17.311 17.509 0.4509 0.3398 1.1356 1.1601 0.8091
Im10 16.768 16.137 16.537 16.639 16.775 0.4950 0.3888 1.2307 1.2749 0.8803
Im1 20.772 19.855 20.440 20.498 20.820 0.6297 0.4993 1.4758 1.6274 1.1303
Im2 20.777 19.978 20.377 20.476 20.818 0.6217 0.4843 1.4581 1.5907 1.1056
Im3 20.443 19.633 20.131 20.181 20.454 0.6104 0.4760 1.4386 1.5877 1.0992
Im4 20.914 20.058 20.556 20.642 20.951 0.5965 0.4619 1.4034 1.5422 1.0706
Im5 18.262 17.474 17.925 18.024 18.322 0.6224 0.4824 1.4562 1.6092 1.1210
8
Im6 17.387 15.860 16.876 17.121 17.426 0.6308 0.5039 1.5069 1.6413 1.1361
Im7 20.870 20.910 20.551 20.613 20.005 0.6162 0.4811 1.4593 1.6092 1.1109
Im8 20.874 20.835 20.511 20.608 20.150 0.5147 0.3862 1.3623 1.2830 0.8980
Im9 20.983 21.040 20.689 20.594 20.130 0.5427 0.3921 1.3938 1.3376 0.9385
Im10 19.976 20.017 20.143 19.737 19.962 0.5427 0.3926 1.3639 1.3410 0.9446
Im1 29.391 28.028 28.739 28.733 29.468 0.8410 0.5954 2.1359 2.0634 1.4480
Im2 29.682 28.539 29.102 29.027 29.756 0.8407 0.5974 2.1409 2.0494 1.4431
Im3 29.261 27.703 28.514 28.524 29.262 0.8273 0.5764 2.0536 2.0123 1.4219
Im4 29.531 29.634 28.958 28.912 28.968 0.8236 0.5854 2.0602 1.9981 1.4062
Im5 25.204 25.215 24.485 24.551 24.867 0.8387 0.5844 2.0817 2.0453 1.4407
15
Im6 23.630 23.617 22.565 22.898 22.390 0.8608 0.6066 2.1608 2.0980 1.4844
Im7 29.474 29.597 28.855 28.928 28.814 0.8155 0.5842 2.0720 2.0104 1.4097
Im8 30.066 28.884 29.465 29.557 30.139 0.7050 0.4737 1.9618 1.6714 1.1838
Im9 29.748 30.010 29.143 29.100 28.397 0.7102 0.4692 1.9408 1.6659 1.1879
Im10 28.868 28.295 28.283 28.945 27.402 0.7054 0.4848 1.9741 1.6817 1.1950
Im1 31.958 31.944 31.181 31.123 30.500 0.8932 0.6258 2.2970 2.1670 1.5326
Im2 32.391 31.163 31.741 31.711 32.433 0.8849 0.6256 2.2696 2.1428 1.5152
Im3 31.786 30.081 31.022 30.929 31.791 0.8801 0.6176 2.2763 2.1513 1.5177
Im4 32.134 30.814 31.484 31.456 32.140 0.9005 0.6386 2.2988 2.1663 1.5292
Im5 27.163 27.214 26.436 26.330 26.608 0.8855 0.6263 2.3080 2.1530 1.5172
17
Im6 25.282 25.287 24.091 24.454 24.278 0.8777 0.6186 2.2895 2.1665 1.5232
Im7 32.107 31.378 31.427 32.194 31.800 0.8972 0.6245 2.3090 2.1649 1.5269
Im8 32.678 32.019 32.037 32.710 31.399 0.7798 0.5273 2.1741 1.8256 1.3054
Im9 32.444 31.671 31.660 32.533 31.976 0.7724 0.5175 2.1660 1.8259 1.2881
Im10 31.460 31.581 30.841 30.769 30.944 0.7727 0.5025 2.1519 1.7899 1.2810
Im1 33.523 34.234 33.574 34.364 33.752 0.9539 0.6580 2.4952 2.2919 1.6254
Im2 34.283 34.973 34.224 34.975 34.268 0.9572 0.6575 2.5105 2.2805 1.6134
Im3 34.220 34.140 33.267 33.250 33.289 0.9444 0.6430 2.4728 2.2559 1.5984
Im4 33.058 34.648 34.814 33.857 33.937 0.9495 0.6456 2.4701 2.2631 1.6016
Im5 28.996 29.036 28.173 28.091 28.331 0.9589 0.6650 2.5054 2.2971 1.6227
19
Im6 26.751 26.540 25.479 25.888 25.947 0.9650 0.6843 2.5953 2.3158 1.6435
Im7 34.636 34.730 33.863 33.800 34.077 0.9601 0.6559 2.4579 2.2940 1.6271
Im8 35.203 34.531 34.482 35.228 34.734 0.8373 0.5410 2.3879 1.9500 1.3883
Im9 34.962 34.022 35.787 34.169 34.296 0.9081 0.5836 2.5097 2.1084 1.4999
Im10 33.922 33.134 33.288 34.015 33.536 0.8162 0.5307 2.3427 1.9044 1.3563
Im1 41.070 40.643 40.023 40.007 40.965 1.1320 0.7448 3.1039 2.6453 1.8826
Im2 42.187 41.869 41.262 41.091 41.351 1.1306 0.7578 3.1007 2.6337 1.8725
Im3 40.607 40.248 39.427 39.297 40.241 1.1387 0.7417 3.1092 2.6594 1.8932
Im4 39.773 41.224 40.550 40.511 41.555 1.1279 0.7467 3.0678 2.6322 1.8738
Im5 32.050 33.722 32.975 32.777 33.837 1.1876 0.7839 3.1912 2.7647 1.9736
25
Im6 26.775 29.623 28.746 29.748 30.470 1.1602 0.7610 3.1010 2.7464 1.9457
Im7 41.592 41.486 40.650 40.546 39.780 1.1364 0.7534 3.0817 2.6599 1.8931
Im8 40.341 42.110 42.705 41.281 41.531 1.0198 0.6404 3.0088 2.3044 1.6551
Im9 40.890 41.994 41.126 40.092 41.087 1.1658 0.7380 3.2423 2.5647 1.8657
Im10 39.830 40.496 40.817 39.612 40.268 1.0496 0.6523 3.0418 2.3930 1.7189
Average 27.996 27.824 27.695 27.711 27.752 0.804 0.564 2.092 1.949 1.375

Table 10
Mean rank of each algorithm using Friedman’s test based on the performance measures.
CS GWO SSA FPA FO-FPA
PSNR 1.85 1.38 3.63 3.76 4.36
SSIM 1.7 1.75 3.59 3.7 4.25
Fitness value 3.75 3.11 2.53 2.56 3.03
CPU time(s) 2 1 4.71 4.28 3
28 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889

Fig. A.1. Eighteen threshold values placed over the histogram of Im1–Im6.

1. PSNR [59]: It is defined in Eq . (23) which calculate the

255
difference between image I and its segmented image I seg. PSNR = 20log10 ( ), (in dB) (23)
RMSE
 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889 29

Fig. A.2. Eighteen threshold values placed over the histogram of Img7–Img10.

where M and N are the number of rows and columns in that used in the mathematical validation are used except the
image, while, RMSE is the root mean-squared error and it population size set to 20 and the total number of iterations set
is formulated as: to 50.
The average PSNR values overall the tested images and thresh-
√
∑N ∑M
i=1 j=1 (Iij − I segij )2
RMSE = (24) olds are given in Figs. 5(a)–5(b) and Table 7. From Table 7 we
N .M observed that the results of the proposed FO-FPA at the different
2. SSIM [60]: It is defined in Eq. (25) which compute the value of the parameters are competitive. However, it can be
similarity between I and I seg. noticed that the performance of the FO-FPA at α = 0.6 and r = 8
is better than others. For example, by considering there are thirty
(2µI µI seg + c1 )(2σI ,I seg + c2 )
SSIM(I , I seg) = , cases (three threshold levels × ten images), the FO-FPA at these
(µ2I + µ2I seg + c1 )(σI1 + σI2seg + c2 ) values for α = 0.6 and r = 8 has the largest average of PSNR
c1 = 6.5025, c2 = 58.52252 among all other values. Moreover, from Fig. 5(a), that depicts the
(25) PSNR’s average at each threshold, it can be seen that the better
value for α and r is 0.6 and 8, respectively, at the three levels
In Eq. (25), the µI (µI seg ) and σI (σI seg ) are the mean inten-
followed by (α = 0.6 and r = 4) at level 2, while, at (α = 0.4
sity and the standard deviation of I(I seg), respectively. The
and r = 4), and (α = 0.4 and r = 12) at levels 12, and 28,
σI ,Seg is the covariance of I seg and I.
respectively. Also, this indicates that the r = 8 supports the FO-
4.2.4. Determine best configuration FPA to provide better results than other terms (i.e, 4, and 12).
In this experimental, the main aim is to determine the best In addition, when we used the rank to determine the optimal
configuration from the parameters of FPA. The proposed FO-FPA configuration from the parameters of FO-FPA, it can be noticed
is evaluated at different three r (4, 8, and 12), and α (0.4, 0.5, from Fig. 5(b) that the α = 0.6 is the more suitable value for the
and 0.6). In addition to, a set of three threshold values (i.e., 2, 12, FO-FPA, followed by either α = 0.4 (with r either 4 or 12) or
28) and ten images. The same parameter of the two algorithms α = 0.6 (with r = 4) since both cases have the same rank.
30 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889

Fig. A.3. Segmented images using eighteen Threshold values for Im1–Im6.

4.2.5. Comparison with metaheuristic methods is performed at different threshold levels (i.e., 6, 8, 15, 17, 19, and
To show the effectiveness of the proposed FO-FPA as an image 25) as given in Tables 8–9 and Figs. 6–A.4.
segmentation method, it is compared with a set of algorithms Table 8 illustrates the PSNR and SSIM values for each algo-
such as CS, GWO, SSA, and FPA. Since, these algorithm established rithm and among the tested images and threshold values. Since
as global optimization method in Section 4.1.3. This comparison there are ten images and six threshold values, we have sixteen
 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889 31

Fig. A.4. Segmented images using eighteen Threshold values for Img7–Img10.

cases in total. Therefore, the results in Table 8 sheds lights on cases from sixteen cases (i.e., 31%), followed by GWO and CS
the high quality of the proposed FO-FPA. Firstly, the GWO and which allocates the second and third rank, respectively, with
CS are the worst algorithms, in this study, according to the PSNR fourteen (23%) and eleven (18%) cases. In addition, the SSA and
values, while, the SSA provides better results than both of them FPA have the same number of cases (i.e., eight cases). In general,
(i.e., CS and GWO). By comparing the results of FO-FPA and FPA from the average overall the cases it can be noticed that the best
is can be seen that the FO-FPA achieves the largest PSNR value algorithm in term of fitness value is the CS followed by GWO,
over thirty-seven cases (i.e., 61% from the cases), whereas the while, the FO-FPA allocates the third rank which provide fitness
FPA allocates the second best rank with nineteen cases (i.e., 31% value better than SSA and FPA. But the all the algorithms nearly
from the cases). Secondly, the SSIM value of the FO-FPA is higher have the same fitness value with little difference and this can be
than other algorithms in the most of the cases nearly thirty-eight observed, also, from Fig. 8.
cases (63%), followed by the FPA which has the best SSIM value As per the CPU time(s) given in Fig. 9, it can be seen that the
at sixteen cases (26%). The SSA and GWO achieve the third and CS and GWO are the two fast algorithms, in this study, which
fourth rank with four and two cases, respectively, while, the CS allocate the first and second rank. Whereas, the proposed FO-FPA
still the worst case. is the third fast algorithm which takes time less than the other
In addition, Figs. 6 and 7 depict the average of PSNR and SSIM two algorithm FPA and SSA.
of each algorithm along each threshold level, respectively. From Figs. A.1–A.2 show the eighteen threshold values obtained by
these figures it can be noticed that the proposed FO-FPA provides each algorithms which placed over the histogram of each images.
better results especially at the threshold level 6, 8 , and 19, while In addition to, Figs. A.3–A.4 show the segmented images at the
its results at the rest threshold values are competitive with other same level (i.e., eighteen). From these figures it can be observed
algorithms. However, the average of the proposed FO-FPA overall the quality of the segmented images according to the threshold
the tested thresholds and images is the best in terms of PSNR and obtained by FO-FPA.
SSIM values.
Table 9 and Fig. 8 show the results of the algorithms in terms 4.2.6. Friedman’s test
of fitness value (i.e., fuzzy entropy). From these results it can be In this section, we applied the Friedman’s test to determine if
observed that the FO-FPA has the higher fitness value at nineteen there is a significant difference between the FO-FPA and other
32 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889

methods that used to find the threshold values. This test is [7] S. Mirjalili, S.M. Mirjalili, A. Lewis, Grey wolf optimizer, Adv. Eng. Softw.
applied for the results overall the tested images and threshold 69 (2014) 46–61.
values. [8] A.H. Gandomi, X.-S. Yang, A.H. Alavi, Cuckoo search algorithm: a meta-
heuristic approach to solve structural optimization problems, Eng. Comput.
Table 10 shows the mean rank for each algorithm and it can 29 (1) (2013) 17–35.
be concluded from this table that the FO-FPA achieves the first [9] S. Mirjalili, A.H. Gandomi, S.Z. Mirjalili, S. Saremi, H. Faris, S.M. Mirjalili,
rank in terms of PSNR and SSIM. Meanwhile, the CS and GWO Salp swarm algorithm: A bio-inspired optimizer for engineering design
achieve the best rank in terms of fitness value and CPU time(s), problems, Adv. Eng. Softw. 114 (2017) 163–191.
respectively. [10] H. Jouhari, D. Lei, M.A. Al-qaness, M.A. Elaziz, A.A. Ewees, O. Farouk, Sine-
cosine algorithm to enhance simulated annealing for unrelated parallel
machine scheduling with setup times, Mathematics 7 (11) (2019) 1120.
5. Conclusion [11] M.A. Elaziz, A.A. Ewees, R.A. Ibrahim, S. Lu, Opposition-based moth-flame
optimization improved by differential evolution for feature selection, Math.
In this paper, we improved the performance of the FPA algo- Comput. Simulation 168 (2020) 48–75.
rithm by using the concept of fractional-order (FO) calculus. The [12] D. Yousri, T.S. Babu, D. Allam, V. Ramachandaramurthy, E. Beshr, M. Eteiba,
et al., Fractional chaos maps with flower pollination algorithm for partial
FO calculus memory property enhanced the FPA with a good tool shading mitigation of photovoltaic systems, Energies 12 (18) (2019) 3548.
which improved its local search ability and provided an adaptive [13] D. Yousri, T.S. Babu, D. Allam, V.K. Ramachandaramurthy, M.B. Etiba, A
updating for the probability of switching between its exploration novel chaotic flower pollination algorithm for global maximum power
and exploitation. To check the quality of the FO-FPA approach, we point tracking for photovoltaic system under partial shading conditions,
applied it to solve different kind of problems through performing IEEE Access 7 (2019) 121432–121445.
[14] S. Mirjalili, A.H. Gandomi, Chaotic gravitational constants for the
two experiments series. The aim of the first experiment series gravitational search algorithm, Appl. Soft Comput. 53 (2017) 407–419.
was to evaluate the ability of the FO-FPA to solve thirty-six func- [15] D. Alam, D. Yousri, M. Eteiba, Flower pollination algorithm based solar pv
tions that are different in complexity and dimensions. Whereas, parameter estimation, Energy Convers. Manage. 101 (2015) 410–422.
the second experimental series attempted to apply the proposed [16] M. Samy, S. Barakat, H. Ramadan, A flower pollination optimization
FO-FPA to a real application, and this achieved by implementing algorithm for an off-grid pv-fuel cell hybrid renewable system, Int. J.
Hydrog. Energy 44 (4) (2019) 2141–2152.
it as a multi-level image segmentation method to determine the
[17] S. Arora, P. Anand, Chaos-enhanced flower pollination algorithms for global
thresholds that will divide the image into a set of classes. In this optimization, J. Intell. Fuzzy Systems 33 (6) (2017) 3853–3869.
application, the fuzzy entropy is used to assess the quality of the [18] R.O. Abdel, B.M. Abdel, I. El Henawy, A new hybrid flower pollination
solution during the optimization process. According to the results algorithm for solving constrained global optimization problems, 2014.
of the two experiments, the proposed FO-FPA outperforms the [19] R. Wang, Y. Zhou, S. Qiao, K. Huang, Flower pollination algorithm with bee
pollinator for cluster analysis, Inform. Process. Lett. 116 (1) (2016) 1–14.
other MH methods such as GWO, CS, SSA, and traditional FPA.
[20] L. Valenzuela, F. Valdez, P. Melin, Flower pollination algorithm with fuzzy
Therefore, the proposed FO-FPA can be employed in other approach for solving optimization problems, in: Nature-Inspired Design of
applications in future works such as automatic data clustering, Hybrid Intelligent Systems, Springer, 2017, pp. 357–369.
feature selection, and task scheduling in cloud computing. [21] H.M. Dubey, M. Pandit, B.K. Panigrahi, A biologically inspired modified
flower pollination algorithm for solving economic dispatch problems in
modern power systems, Cogn. Comput. 7 (5) (2015) 594–608.
Declaration of competing interest
[22] D. Chakraborty, S. Saha, O. Dutta, De-fpa: a hybrid differential evolution-
flower pollination algorithm for function minimization, in: 2014 Inter-
The authors declare that they have no known competing finan- national Conference on High Performance Computing and Applications
cial interests or personal relationships that could have appeared (ICHPCA), IEEE, 2014, pp. 1–6.
to influence the work reported in this paper. [23] E. Nabil, A modified flower pollination algorithm for global optimization,
Expert Syst. Appl. 57 (2016) 192–203.
[24] R. Wang, Y. Zhou, Flower pollination algorithm with dimension by
CRediT authorship contribution statement dimension improvement, Math. Probl. Eng. 2014 (2014).
[25] W. Yamany, H.M. Zawbaa, E. Emary, A.E. Hassanien, Attribute reduction
Dalia Yousri: Conceptualization, Methodology, Software, Data approach based on modified flower pollination algorithm, in: 2015 IEEE
curation, Visualization, Formal analysis, Writing - review & edit- International Conference on Fuzzy Systems (FUZZ-IEEE), IEEE, 2015, pp.
ing. Mohamed Abd Elaziz: Conceptualization, Methodology, Soft- 1–7.
[26] Y. Zhou, R. Wang, An improved flower pollination algorithm for opti-
ware, Data curation, Visualization, Formal analysis, Writing - mal unmanned undersea vehicle path planning problem, Int. J. Pattern
review & editing. Seyedali Mirjalili: Conceptualization, Software, Recognit. Artif. Intell. 30 (04) (2016) 1659010.
Supervision, Project administration, Writing- review & editing. [27] P. Niu, J. Li, L. Chang, X. Zhang, R. Wang, G. Li, A novel flower pollination
algorithm for modeling the boiler thermal efficiency, Neural Process. Lett.
Appendix 49 (2) (2019) 737–759.
[28] M.K.Y. Shambour, A.A. Abusnaina, A.I. Alsalibi, Modified global flower polli-
nation algorithm and its application for optimization problems, Interdiscip.
See Figs. A.1–A.4. Sci., Comput. Life Sci. 11 (3) (2019) 496–507.
[29] O. Abdel-Raouf, I. El-Henawy, M. Abdel-Baset, et al., A novel hybrid flower
References pollination algorithm with chaotic harmony search for solving sudoku
puzzles, Int. J. Mod. Educ. Comput. Sci. 6 (3) (2014) 38.
[1] N. Neggaz, A.A. Ewees, M.A. Elaziz, M. Mafarja, Boosting salp swarm [30] S. Kalra, S. Arora, Firefly algorithm hybridized with flower pollination
algorithm by Sine cosine algorithm and disrupt operator for feature algorithm for multimodal functions, in: Proceedings of the International
selection, Expert Syst. Appl. (2019) 113103. Congress on Information and Communication Technology, Springer, 2016,
[2] J.R. Koza, Genetic Programming, Citeseer, 1997. pp. 207–219.
[3] F.A. Hashim, E.H. Houssein, M.S. Mabrouk, W. Al-Atabany, S. Mirjalili, Henry [31] R. Salgotra, U. Singh, A novel bat flower pollination algorithm for synthesis
gas solubility optimization: A novel physics-based algorithm, Future Gener. of linear antenna arrays, Neural Comput. Appl. 30 (7) (2018) 2269–2282.
Comput. Syst. 101 (2019) 646–667. [32] D. Rodrigues, G.H. de Rosa, L.A. Passos, J.P. Papa, Adaptive improved
[4] S. Mirjalili, S.M. Mirjalili, A. Hatamlou, Multi-verse optimizer: a nature- flower pollination algorithm for global optimization, in: Nature-Inspired
inspired algorithm for global optimization, Neural Comput. Appl. 27 (2) Computation in Data Mining and Machine Learning, Springer, 2020, pp.
(2016) 495–513. 1–21.
[5] E. Rashedi, H. Nezamabadi-Pour, S. Saryazdi, Gsa: a gravitational search [33] D.H. Wolpert, W.G. Macready, No free lunch theorems for optimization,
algorithm, Inform. Sci. 179 (13) (2009) 2232–2248. IEEE Trans. Evol. Comput. 1 (1) (1997) 67–82.
[6] I.C. Trelea, The particle swarm optimization algorithm: convergence [34] Y. Mousavi, A. Alfi, Fractional calculus-based firefly algorithm applied
analysis and parameter selection, Inform. Process. Lett. 85 (6) (2003) to parameter estimation of chaotic systems, Chaos Solitons Fractals 114
317–325. (2018) 202–215.
 D. Yousri, M.A. Elaziz and S. Mirjalili / Knowledge-Based Systems 197 (2020) 105889 33

[35] M.S. Couceiro, R.P. Rocha, N.F. Ferreira, J.T. Machado, Introducing the [48] X. Yu, X. Ye, Q. Gao, Pipeline image segmentation algorithm and heat loss
fractional-order darwinian pso, Signal, Image Video Process. 6 (3) (2012) calculation based on gene-regulated apoptosis mechanism, Int. J. Press.
343–350. Vessels Pip. 172 (2019) 329–336.
[36] E.S. Pires, J.T. Machado, P. de Moura Oliveira, J.B. Cunha, L. Mendes, Particle [49] D. Oliva, S. Nag, M.A. Elaziz, U. Sarkar, S. Hinojosa, Multilevel thresholding
swarm optimization with fractional-order velocity, Nonlinear Dynam. 61 by fuzzy type II sets using evolutionary algorithms, Swarm Evol. Comput.
(1–2) (2010) 295–301. 51 (2019) 100591.
[37] A.B. Deshmukh, N.U. Rani, Fractional-grey wolf optimizer-based kernel [50] M.A. Elaziz, D. Oliva, A.A. Ewees, S. Xiong, Multi-level thresholding-
weighted regression model for multi-view face video super resolution, Int. based grey scale image segmentation using multi-objective multi-verse
J. Mach. Learn. Cybern. (2017) 1–19. optimizer, Expert Syst. Appl. 125 (2019) 112–129.
[38] X.-S. Yang, Flower pollination algorithm for global optimization, in: [51] O. Tarkhaneh, H. Shen, An adaptive differential evolution algorithm to
International Conference on Unconventional Computing and Natural optimal multi-level thresholding for mri brain image segmentation, Expert
Computation, Springer, 2012, pp. 240–249. Syst. Appl. 138 (2019) 112820.
[39] D. Yousri, A.M. AbdelAty, A.G. Radwan, A.S. Elwakil, C. Psychalinos, [52] X. Liu, X. Zhu, M. Li, L. Wang, E. Zhu, T. Liu, M. Kloft, D. Shen, J. Yin, W.
Comprehensive comparison based on meta-heuristic algorithms for ap- Gao, Multiple kernel k-means with incomplete kernels, IEEE Trans. Pattern
proximation of the fractional-order Laplacian sα as a weighted sum of Anal. Mach. Intell. (2019).
first-order high-pass filters, Microelectron. J. 87 (2019) 110–120. [53] K. Hammouche, M. Diaf, P. Siarry, A comparative study of various meta-
[40] I. Podlubny, Fractional Differential Equations: An Introduction to Fractional heuristic techniques applied to the multilevel thresholding problem, Eng.
Derivatives, Fractional Differential Equations, to Methods of their Solution Appl. Artif. Intell. 23 (5) (2010) 676–688.
and Some of their Applications, Vol. 198, Elsevier, 1998. [54] M.A. El Aziz, A.A. Ewees, A.E. Hassanien, Whale optimization algorithm and
[41] P. Ostalczyk, Discrete Fractional Calculus: Applications in Control and moth-flame optimization for multilevel thresholding image segmentation,
Image Processing, Vol. 4, World Scientific, 2015. Expert Syst. Appl. 83 (2017) 242–256.
[42] M.A. Elaziz, S. Mirjalili, A hyper-heuristic for improving the initial pop- [55] K.B. Resma, M.S. Nair, Multilevel thresholding for image segmentation
ulation of whale optimization algorithm, Knowl.-Based Syst. 172 (2019) using krill herd optimization algorithm, J. King Saud Univ.-Comput. Inform.
42–63. Sci. (2018).
[43] F. Wilcoxon, S. Katti, R.A. Wilcox, Critical values and probability levels for [56] S.C. Satapathy, N.S.M. Raja, V. Rajinikanth, A.S. Ashour, N. Dey, Multi-level
the Wilcoxon rank sum test and the wilcoxon signed rank test, in: Selected image thresholding using otsu and chaotic bat algorithm, Neural Comput.
Tables in Mathematical Statistics, Vol. 1, 1970, pp. 171–259. Appl. 29 (12) (2018) 1285–1307.
[44] N.M. Zaitoun, M.J. Aqel, Survey on image segmentation techniques, [57] A. Raj, G. Gautam, S.N.H.S. Abdullah, A.S. Zaini, S. Mukhopadhyay, Multi-
Procedia Comput. Sci. 65 (2015) 797–806. level thresholding based on differential evolution and Tsallis fuzzy entropy,
[45] S. Amerifar, A.T. Targhi, M.M. Dehshibi, Iris the picture of health: Towards Image Vis. Comput. 91 (2019) 103792.
medical diagnosis of diseases based on iris pattern, in: 2015 Tenth Interna- [58] D. Oliva, M.A. Elaziz, S. Hinojosa, Multilevel thresholding for image seg-
tional Conference on Digital Information Management (ICDIM), IEEE, 2015, mentation based on metaheuristic algorithms, in: Metaheuristic Algorithms
pp. 120–123. for Image Segmentation: Theory and Applications, Springer, 2019, pp.
[46] A.K. Bhandari, A. Kumar, G.K. Singh, Tsallis entropy based multilevel 59–69.
thresholding for colored satellite image segmentation using evolutionary [59] P.I. Yin, Multilevel minimum cross entropy threshold selection based on
algorithms, Expert Syst. Appl. 42 (22) (2015) 8707–8730. particle swarm optimization., Appl. Math. Comput. 184 (2007) 503–892.
[47] X. Yu, Blurred trace infrared image segmentation based on template [60] Z. Wang, A.C. Bovik, H.R. Sheikh, E.P. Simoncelli, Image quality assess-
approach and immune factor, Infrared Phys. Technol. 67 (2014) 116–120. ment: From error measurement to structural similarity, IEEE Trans. Image
Process. 13 (1) (2004).