catswarmoptimizationalgorithm:asurveyand …downloads.hindawi.com/journals/cin/2020/4854895.pdf ·...

Review ArticleCat Swarm Optimization Algorithm A Survey andPerformance Evaluation

Aram M Ahmed 12 Tarik A Rashid 3 and Soran Ab M Saeed2

1International Academic Office Kurdistan Institution for Strategic Studies and Scientific Research Sulaymaniyah 46001 Iraq2Information Technology Sulaimani Polytechnic University Sulaymaniyah 46001 Iraq3Computer Science and Engineering University of Kurdistan Hewler Erbil 44001 Iraq

Correspondence should be addressed to Aram M Ahmed aramahmedkissredukrd

Received 25 July 2019 Revised 15 December 2019 Accepted 20 December 2019 Published 22 January 2020

Academic Editor Juan A Gomez-Pulido

Copyright copy 2020 Aram M Ahmed et al +is is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited

+is paper presents an in-depth survey and performance evaluation of cat swarm optimization (CSO) algorithm CSO is a robustand powerful metaheuristic swarm-based optimization approach that has received very positive feedback since its emergence Ithas been tackling many optimization problems and many variants of it have been introduced However the literature lacks adetailed survey or a performance evaluation in this regard +erefore this paper is an attempt to review all these works includingits developments and applications and group them accordingly In addition CSO is tested on 23 classical benchmark functionsand 10 modern benchmark functions (CEC 2019) +e results are then compared against three novel and powerful optimizationalgorithms namely dragonfly algorithm (DA) butterfly optimization algorithm (BOA) and fitness dependent optimizer (FDO)+ese algorithms are then ranked according to Friedman test and the results show that CSO ranks first on the whole Finallystatistical approaches are employed to further confirm the outperformance of CSO algorithm

1 Introduction

Optimization is the process by which the optimal solution isselected for a given problem among many alternative so-lutions One key issue of this process is the immensity of thesearch space for many real-life problems in which it is notfeasible for all solutions to be checked in a reasonable timeNature-inspired algorithms are stochastic methods whichare designed to tackle these types of optimization problems+ey usually integrate some deterministic and randomnesstechniques together and then iteratively compare a numberof solutions until a satisfactory one is found +ese algo-rithms can be categorized into trajectory-based and pop-ulation-based classes [1] In trajectory-based types such as asimulated annealing algorithm [2] only one agent issearching in the search space to find the optimal solutionwhereas in the population-based algorithms also known asswarm Intelligence such as particle swarm optimization(PSO) [3] multiple agents are searching and communicating

with each other in a decentralized manner to find the op-timal solution Agents usually move in two phases namelyexploration and exploitation In the first one they move on aglobal scale to find promising areas while in the second onethey search locally to discover better solutions in thosepromising areas found so far Having a trade-off betweenthese two phases in any algorithm is very crucial becausebiasing towards either exploration or exploitation woulddegrade the overall performance and produce undesirableresults [1] +erefore more than hundreds of swarm in-telligence algorithms have been proposed by researchers toachieve this balance and provide better solutions for theexisting optimization problems

Cat swarm optimization (CSO) is a swarm Intelligencealgorithm which was originally invented by Chu et al in2006 [4 5] It is inspired by the natural behavior of cats andit has a novel technique in modeling exploration and ex-ploitation phases It has been successfully applied in variousoptimization fields of science and engineering However the

HindawiComputational Intelligence and NeuroscienceVolume 2020 Article ID 4854895 20 pageshttpsdoiorg10115520204854895

literature lacks a recent and detailed review of this algorithmIn addition since 2006 CSO has not been compared againstnovel algorithms ie it has been mostly compared with PSOalgorithm while many new algorithms have been introducedsince then So a question which arises is whether CSOcompetes with the novel algorithms or not +ereforeexperimenting CSO on a wider range of test functions andcomparing it with new and robust algorithms will furtherreveal the potential of the algorithm As a result the aims ofthis paper are as follows firstly provide a comprehensiveand detailed review of the state of art of CSO algorithm (seeFigure 1) which shows the general framework for con-ducting the survey secondly evaluate the performance ofCSO algorithm against modern metaheuristic algorithms+ese should hugely help researchers to further work in thedomain in terms of developments and applications

+e rest of the paper is organized as follows Section 2presents the original algorithm and its mathematicalmodeling Section 3 is dedicated to reviewing all modifiedversions and variants of CSO Section 4 summarizes thehybridizing CSO algorithm with ANN and other non-metaheuristic methods Section 5 presents applications ofthe algorithm and groups them according to their disci-plinary Section 6 provides performance evaluation whereCSO is compared against dragonfly algorithm (DA) [6]butterfly optimization algorithm (BOA) [7] and fitnessdependent optimizer (FDO) [8] Finally Section 7 providesthe conclusion and future directions

2 Original Cat Swarm Optimization Algorithm

+e original cat swarm optimization is a continuous andsingle-objective algorithm [4 5] It is inspired by resting andtracing behaviours of cats Cats seem to be lazy and spendmost of their time resting However during their rests theirconsciousness is very high and they are very aware of what ishappening around them So they are constantly observingthe surroundings intelligently and deliberately and whenthey see a target they start moving towards it quickly+erefore CSO algorithm is modeled based on combiningthese two main deportments of cats

CSO algorithm is composed of two modes namelytracing and seeking modes Each cat represents a solutionset which has its own position a fitness value and a flag+eposition is made up of M dimensions in the search spaceand each dimension has its own velocity the fitness valuedepicts how well the solution set (cat) is finally the flag is toclassify the cats into either seeking or tracing mode+us weshould first specify how many cats should be engaged in theiteration and run them through the algorithm +e best catin each iteration is saved into memory and the one at thefinal iteration will represent the final solution

21 General Structure of the Algorithms +e algorithm takesthe following steps in order to search for optimal solutions

(1) Specify the upper and lower bounds for the solutionsets

(2) Randomly generate N cats (solution sets) and spreadthem in the M dimensional space in which each cathas a random velocity value not larger than a pre-defined maximum velocity value

(3) Randomly classify the cats into seeking and tracingmodes according to MR MR is a mixture ratiowhich is chosen in the interval of [0 1] So forexample if a number of cats N is equal to 10 and MRis set to 02 then 8 cats will be randomly chosen to gothrough seeking mode and the other 2 cats will gothrough tracing mode

(4) Evaluate the fitness value of all the cats according tothe domain-specified fitness function Next the bestcat is chosen and saved into memory

(5) +e cats thenmove to either seeking or tracingmode(6) After the cats go through seeking or tracing mode

for the next iteration randomly redistribute the catsinto seeking or tracing modes based on MR

(7) Check the termination condition if satisfied termi-nate the program otherwise repeat Step 4 to Step 6

22 Seeking Mode +is mode imitates the resting behaviorof cats where four fundamental parameters play importantroles seeking memory pool (SMP) seeking range of theselected dimension (SRD) counts of dimension to change(CDC) and self-position considering (SPC) +ese valuesare all tuned and defined by the user through a trial-and-error method

SMP specifies the size of seeking memory for cats ie itdefines number of candidate positions in which one of themis going to be chosen by the cat to go to for example if SMPwas set to 5 then for each and every cat 5 new randompositions will be generated and one of them will be selectedto be the next position of the cat How to randomize the newpositions will depend on the other two parameters that areCDC and SRD CDC defines how many dimensions to bemodified which is in the interval of [0 1] For example if thesearch space has 5 dimensions and CDC is set to 02 then foreach cat four random dimensions out of the five need to bemodified and the other one stays the same SRD is themutative ratio for the selected dimensions ie it defines theamount of mutation and modifications for those dimensionsthat were selected by CDC Finally SPC is a Boolean valuewhich specifies whether the current position of a cat will

CSO and its variants with artificial neural

networks

Original CSO algorithm Variants of CSO

Applications of CSO

Figure 1 General framework for conducting the survey

2 Computational Intelligence and Neuroscience

be selected as a candidate position for the next iteration ornot So for example if the SPC flag is set to true then foreach cat we need to generate (SMP-1) number of can-didates instead of SMP number as the current position isconsidered as one of them Seeking mode steps are asfollows

(1) Make as many as SMP copies of the current positionof Catk

(2) For each copy randomly select as many as CDCdimensions to be mutated Moreover randomly addor subtract SRD values from the current valueswhich replace the old positions as shown in thefollowing equation

Xjdnew (1 + randlowast SRD)lowastXjdold (1)

where Xjdold is the current position Xjdnew is thenext position j denotes the number of a cat and ddenotes the dimensions and rand is a randomnumber in the interval of [0 1]

(3) Evaluate the fitness value (FS) for all the candidatepositions

(4) Based on probability select one of the candidatepoints to be the next position for the cat wherecandidate points with higher FS have more chanceto be selected as shown in equation (2) Howeverif all fitness values are equal then set all theselecting probability of each candidate point tobe 1

Pi FSi minus FSb

11138681113868111386811138681113868111386811138681113868

FSmax minus FSmin where 0lt ilt j (2)

If the objective is minimization then FSb FSmax oth-erwise FSb FSmin

23 TracingMode +is mode copies the tracing behavior ofcats For the first iteration random velocity values are givento all dimensions of a catrsquos position However for later stepsvelocity values need to be updated Moving cats in this modeare as follows

(1) Update velocities (Vkd) for all dimensions accordingto equation (3)

(2) If a velocity value outranged the maximum valuethen it is equal to the maximum velocity

Vkd Vkd + r1c1 Xbestd minus Xkd1113872 1113873 (3)

(3) Update position of Catk according to the followingequation

Xkd Xkd + Vkd (4)

Refer to Figure 2 which recaps the whole algorithm in adiagram

3 Variants of CSO

In the previous section the original CSO was covered thissection briefly discusses all other variants of CSO found in theliterature Variantsmay include the following points binary ormultiobjective versions of the algorithm changing parame-ters altering steps modifying the structure of the algorithmor hybridizing it with other algorithms Refer to Table 1 whichpresents a summary of these modifications and their results

31 Discrete Binary Cat Swarm Optimization Algorithm(BCSO) Sharafi et al introduced the BCSO Algorithmwhich is the binary version of CSO [9] In the seeking modethe SRD parameter has been substituted by another pa-rameter called the probability of mutation operation (PMO)However the proceeding steps of seeking mode and theother three parameters stay the same Accordingly the di-mensions are selected using the CDC and then PMO will beapplied In the tracing mode the calculations of velocity andposition equations have also been changed into a new formin which the new position vector is composed of binarydigits taken from either current position vector or globalposition vector (best position vector) Two velocity vectorsare also defined in order to decide which vector (current orglobal) to choose from

32 Multiobjective Cat Swarm Optimization (MOCSO)Pradhan and Panda proposed multiobjective cat swarmoptimization (MOCSO) by extending CSO to deal withmultiobjective problems [10] MOCSO is combined with theconcept of the external archive and Pareto dominance inorder to handle the nondominated solutions

33 Parallel Cat Swarm Optimization (PCSO) Tsai and panintroduced parallel cat swarm optimization (PCSO) [11]+is algorithm improved the CSO algorithm by eliminatingthe worst solutions To achieve this they first distribute thecats into subgroups ie subpopulations Cats in the seekingmode move as they do in the original algorithm However inthe tracing mode for each subgroup the best cat will besaved into memory and will be considered as the local bestFurthermore cats move towards the local best rather thanthe global best +en in each group the cats are sortedaccording to their fitness function from best to worst +isprocedure will continue for a number of iterations which isspecified by a parameter called ECH (a threshold that defineswhen to exchange the information of groups) For exampleif ECH was equal to 20 then once every 20 iterations thesubgroups exchange information where the worst cats willbe replaced by a randomly chosen local best of anothergroup +ese modifications lead the algorithm to be com-putationally faster and show more accuracy when thenumber of iteration is fewer and the population size is small

34CSOClustering Santosa andNingrum improved the CSOalgorithm and applied it for clustering purposes [12]+emain

Computational Intelligence and Neuroscience 3

goal was to use CSO to cluster the data and find the best clustercenter +e modifications they did were two main pointsfirstly removing the mixture ratio (MR) and hence forcing allthe cats to go through both seeking and tracing mode +is is

aimed at shortening the time required to find the best clustercenter Secondly always setting the CDC value to be 100instead of 80 as in the original CSO in order to change alldimensions of the candidate cats and increase diversity

Start

Generate N cats

Initialize the position velocitiesand the flag of every cat

Evaluate the cats based on their fitness function

and save the best catinto memory

Catk is in theseeking mode

YesNoSeeking mode

SPC flag is trueYes Make SMP-1

copies of the catNoMake SMP

copies of the cat

Add or substract SRD valuesbased on CDC

Evaluate fitness value ofcats

Based on probability selectnext position for the cat

Tracing mode

Velocity valuegt

maximum value

Yes Velocity = maximumvelocity

No

Update velocitiesfor all dimensions

Updatepositions

Redistribute cats into seeking ortracing modes according to MR

Terminate

Stop

Figure 2 Cat swarm optimization algorithm general structure


Table 1 Summary of the modified versions of the CSO algorithm

Comparison of With Testing field Performance ReferenceCSO (original) PSO and weighted-PSO Six test functions Better [4 5]

BCSO GA BPSO and NBPSOFour test functions (sphere

Rastrigin Ackley andRosenbrock)

Better [9]

MOCSO NSGA-II Cooperative spectrum sensing incognitive radio Better [10]

PCSO CSO and weighted-PSO +ree test functions (RosenbrockRastrigrin and Griewank)

Better when the number ofiteration is fewer and thepopulation size is small

[11]

CSO clustering K-means and PSO clusteringFour different clustering datasets(Iris Soybean Glass and Balance

Scale)More accurate but slower [12]

EPCSOPCSO PSO-LDIW PSO-CREVGCPSO MPSO-TVAC CPSO-

H6 PSO-DVM

Five test functions and aircraftschedule recovery problem Better [13]

AICSO CSO +ree test functions (RastrigrinGriewank and Ackley) Better [14]

ADCSO CSOSix test functions (RastrigrinGriewank Ackley axis parallel

Trid10 and Zakharov)

Better except for Griewank testfunction [15]

Enhanced HCSO PSO Motion estimation block-matching Better [16 17]

ICSO PSO Motion estimation block-matching Better [17]

OL-ICSO K-median PSO CSO and ICSO ART1 ART2 Iris CMC Cancerand Wine datasets Better [18]

CQCSO QCSO CSO PSO and CPSO

Five test functions (SchafferShubert Griewank Rastrigrinand Rosenbrock) and multipeakmaximum power point trackingfor a photovoltaic array under

complex conditions

Better [19]

ICSO CSO and PSO +e 69-bus test distributionsystem Better [20]

ICSO CSO BCSO AICSO and EPCSO

Twelve test functions (sphereRosenbrock Rastrigin GriewankAckley Step Powell Schwefel

Schaffer ZakharovrsquosMichalewicz quartic) and five

real-life clustering problems (IrisCancer CMC Wine and Glass)

Better [21]

Hybrid PCSOABC PCSO and ABC Five test functions Better [22]

CSO-GA-PSOSVM CSO+ SVM (CSOSVM)66 feature points from each face ofCK+ (Cohn Kanade) dataset Better [23]

Hybrid CSO-basedalgorithm GA EA SA PSO and AFS School timetabling test instances Better [24]

Hybrid CSO-GA-SA SLPA and CFinder

Seven datasets (Karate DolphinPolbooks Football Net-Science

Power Indian Railway)Better [25]

MCSO CSO Nine datasets from UCI Better [26]MCSO CSO Eight dataset Better [27]NMCSO CSO PSO Sixteen benchmark functions Better [28]ICSO CSO Ten datasets from UCI Better [29]cCSO DE PSO CSO 47 benchmark functions Better [30]

BBCSO

Binary particle swarmoptimization (BPSO) binary

genetic algorithm (BGA) binaryCSO

01 Knapsack optimizationproblem Better [31]

CSO-CS NA VRP instances from httpneolccumaesvrp NA [32]


35 Enhanced Parallel Cat Swarm Optimization (EPCSO)Tsai et al further improved the PCSO Algorithm in terms ofaccuracy and performance by utilizing the orthogonal arrayof Taguchi method and called it enhanced parallel cat swarmoptimization (EPCSO) [13] Taguchi methods are statisticalmethods which are invented by Japanese Engineer GenichiTaguchi +e idea is developed based on ldquoORTHOGONALARRAYrdquo experiments which improves the engineeringproductivity in thematters of cost quality and performanceIn their proposed algorithm the seeking mode of EPCSO isthe same as the original CSO However the tracingmode hasadopted the Taguchi orthogonal array +e aim of this is toimprove the computational cost even when the number ofagents increases +erefore two sets of candidate velocitieswill be created in the tracing mode +en based on theorthogonal array the experiments will be run and accord-ingly the position of cats will be updated Orouskhani et al[14] added some partial modifications to EPCSO in order tofurther improve it and make it fit their application +emodifications were changing the representation of agentsfrom the coordinate to a set adding a newly defined clusterflag and designing custom-made fitness function

36 Average-Inertia Weighted CSO (AICSO) Orouskhaniet al introduced an inertia value to the velocity equation inorder to achieve a balance between exploration and ex-ploitation phase +ey experimented that (w) value is betterto be selected in the range of [04 09] where at the beginningof the operation it is set to 09 and as the iteration numbermoves forward (w) value gradually becomes smaller until itreaches 04 at the final iteration Large values of (w) assistglobal search whereas small values of (w) assist the localsearch In addition to adding inertia value the positionequation was also reformed to a new one in which averagesof current and previous positions as well as an average ofcurrent and previous velocities were taken in the equation[14]

37 Adaptive Dynamic Cat Swarm Optimization (ADCSO)Orouskhani et al further enhanced the algorithm by in-troducing three main modifications [15] Firstly they in-troduced an adjustable inertia value to the velocity equation+is value gradually decreases as the dimension numbersincrease +erefore it has the largest value for dimensionone and vice versa Secondly they changed the constant (C)to an adjustable value However opposite to the inertiaweight it has the smallest value for dimension one andgradually increases until the final dimension where it has thelargest value Finally they reformed the position equation bytaking advantage of other dimensionsrsquo information

38 Enhanced Hybrid Cat Swarm Optimization (EnhancedHCSO) Hadi and Sabah proposed a hybrid system andcalled it enhanced HCSO [16 17] +e goal was to decreasethe computation cost of the block matching process in videoediting In their proposal they utilized a fitness calculationstrategy in seeking mode of the algorithm +e idea was to

avoid calculating some areas by deciding whether or not todo the calculation or estimate the next search location tomove to In addition they also introduced the inertia weightto the tracing mode

39 Improvement Structure of Cat Swarm Optimization(ICSO) Hadi and Sabah proposed combining two conceptstogether to improve the algorithm and named it ICSO +efirst concept is parallel tracing mode and information ex-changing which was taken from PCSO +e second conceptis the addition of an inertia weight to the position equationwhich was taken from AICSO +ey applied their algorithmfor efficient motion estimation in block matching+eir goalwas to enhance the performance and reduce the number ofiterations without the degradation of the image quality [17]

310 Opposition-Based Learning-Improved CSO (OL-ICSO)Kumar and Sahoo first proposed using Cauchy mutationoperator to improve the exploration phase of the CSO al-gorithm in [34] +en they introduced two more modifi-cations to further improve the algorithm and named itopposition-based learning-improved CSO (OL-ICSO) +eyimproved the population diversity of the algorithm byadopting opposition-based learning method Finally twoheuristic mechanisms (for both seeking and tracing mode)were introduced +e goal of introducing these two mech-anisms was to improve the diverse nature of the populationsand prevent the possibility of falling the algorithm into thelocal optima when the solution lies near the boundary of thedatasets and data vectors cross the boundary constraintsfrequently [18]

311 Chaos Quantum-Behaved Cat Swarm Optimization(CQCSO) Nie et al improved the CSO algorithm in termsof accuracy and avoiding local optima trapping +ey firstintroduced quantum-behaved cat swarm optimization(QCSO) which combined the CSO algorithm with quantummechanics Hence the accuracy was improved and the al-gorithm avoided trapping in the local optima Next byincorporating a tent map technique they proposed chaosquantum-behaved cat swarm optimization (CQCSO) algo-rithm +e idea of adding the tent map was to furtherimprove the algorithm and again let the algorithm to jumpout of the possible local optima points it might fall into [19]

312 Improved Cat Swarm Optimization (ICSO) In theoriginal algorithm cats are randomly selected to either gointo seeking mode or tracing mode using a parameter calledMR However Kanwar et al changed the seeking mode byforcing the current best cat in each iteration to move to theseeking mode Moreover in their problem domain thedecision variables are firm integers while solutions in theoriginal cat are continuous +erefore from selecting thebest cat two more cats are produced by flooring and ceilingits value After that all probable combinations of cats areproduced from these two cats [20]


313 ImprovedCat SwarmOptimization (ICSO) Kumar andSingh made two modifications to the improved CSO algo-rithm and called it ICSO [21] +ey first improved thetracing mode by modifying the velocity and updating po-sition equations In the velocity equation a random uni-formly distributed vector and two adaptive parameters wereadded to tune global and local search movements Secondlya local search method was combined with the algorithm toprevent local optima problem

314Hybrid PCSOABC Tsai et al proposed a hybrid systemby combining PCSO with ABC algorithms and named ishybrid PCSOABC [22] +e structure simply includedrunning PCSO and ABC consecutively Since PCSO per-forms faster with a small population size the algorithm firststarts with a small population and runs PCSO After apredefined number of iterations the population size will beincreased and the ABC algorithm starts running Since theproposed algorithm was simple and did not have any ad-justable feedback parameters it sometimes provided worsesolutions than PCSO Nevertheless its convergence wasfaster than PCSO

315 CSO-GA-PSOSVM Vivek and Reddy proposed a newmethod by combining CSO with particle swarm intelligence(PSO) genetic algorithm (GA) and support vector machine(SVM) and called it CSO-GA-PSOSVM [23] In theirmethod they adopted the GA mutation operator into theseeking mode of CSO in order to obtain divergence Inaddition they adopted all GA operators as well as PSOsubtraction and addition operators into the tracing mode ofCSO in order to obtain convergence +is hybrid meta-heuristic system was then incorporated with the SVMclassifier and applied on facial emotion recognition

316HybridCSO-BasedAlgorithm Skoullis et al introducedthree modifications to the algorithm [24] Firstly theycombined CSO with a local search refining procedureSecondly if the current cat is compared with the global bestcat and their fitness values were the same the global best catwill still be updated by the current cat +e aim of this is toachieve more diversity Finally cats are individually selectedto go into either seeking mode or tracing mode

317 Hybrid CSO-GA-SA Sarswat et al also proposed ahybrid system by combining CSO GA and SA and thenincorporating it with a modularity-based method [25] +eynamed their algorithm hybrid CSO-GA-SA+e structure ofthe system was very simple and straight forward as it wascomposed of a sequential combination of CSO GA and SA+ey applied the system to detect overlapping communitystructures and find near-optimal disjoint communities+erefore input datasets were firstly fed into CSO algorithmfor a predefined number of iterations +e resulted cats werethen converted into chromosomes and henceforth GA wasapplied on them However GA may fall into local optimaand to solve this issue SA was applied afterward

318 Modified Cat Swarm Optimization (MCSO) Lin et alcombined a mutation operator as a local search procedurewith CSO algorithm to find better solutions in the area of theglobal best [26] It is then used to optimize the feature se-lection and parameters of the support vector machineAdditionally Mohapatra et al used the idea of using mu-tation operation before distributing the cats into seeking ortracing modes [27]

319 Normal Mutation Strategy-Based Cat Swarm Optimi-zation (NMCSO) Pappula et al adopted a normal mutationtechnique to CSO algorithm in order to improve the ex-ploration phase of the algorithm +ey used sixteenbenchmark functions to evaluate their proposed algorithmagainst CSO and PSO algorithms [28]

320 Improved Cat Swarm Optimization (ICSO) Lin et alimproved the seeking mode of CSO algorithm Firstly theyused crossover operation to generate candidate positionsSecondly they changed the value of the new position so thatSRD value and current position have no correlations [29] Itis worth mentioning that there are four versions of CSOreferenced in [17 20 21 29] all having the same name(ICSO) However their structures are different

321 Compact Cat Swarm Optimization (CCSO) Zhao in-troduced a compact version of the CSO algorithm A dif-ferential operator was used in the seeking mode of theproposed algorithm to replace the original mutation ap-proach In addition a normal probability model was used inorder to generate new individuals and denote a populationof solutions [30]

322 Boolean Binary Cat Swarm Optimization (BBCSO)Siqueira et al worked on simplifying the binary version ofCSO in order to increase its efficiency +ey reduced thenumber of equations replaced the continues operators withlogic gates and finally integrated the roulette wheel ap-proach with the MR parameter [31]

323HybridCat SwarmOptimization-CrowSearch (CSO-CS)Algorithm Pratiwi proposed a hybrid system by combiningCSO algorithm with crow search (CS) algorithm +e al-gorithm first runs CSO algorithm followed by the memoryupdate technique of the CS algorithm and then new posi-tions will be generated She applied her algorithm on vehiclerouting problem [32]

4 CSO and its Variants with ArtificialNeural Networks

Artificial neural networks are computing systems whichhave countless numbers of applications in various fieldsEarlier neural networks were used to be trained by con-ventional methods such as the backpropagation algorithmHowever current neural networks are trained by nature-


inspired optimization algorithms +e training could beoptimizing the node weights or even the network archi-tectures [35] CSO has also been extensively combined withneural networks in order to be applied in different appli-cation areas +is section briefly goes over those works inwhich CSO is hybridized with ANN and similar methods

41 CSO+ANN+OBD Yusiong proposes combining ANNwith CSO algorithm and optimal brain damage (OBD)approach Firstly the CSO algorithm is used as an opti-mization technique to train the ANN algorithm SecondlyOBD is used as a pruning algorithm to decrease the com-plexity of ANN structure where less number of connectionshas been used As a result an artificial neural network wasobtained that had less training errors and high classificationaccuracy [36]

42 ADCSO+GD+ANFIS Orouskhani et al combinedADCSO algorithm with gradient descent (GD) algorithm inorder to tweak parameters of the adaptive network-basedfuzzy inference system (ANFIS) In their method the an-tecedent and consequent parameters of ANFIS were trainedby CSO algorithm and GD algorithm consecutively [37]

43 CSO+ SVM Abed and Al-Asadi proposed a hybridsystem based on SVM and CSO +e system was applied toelectrocardiograms signals classification +ey used CSO forthe purpose of feature selection optimization and enhancingSVM parameters [38] In addition Lin et al and Wang andWu [39 40] also combined CSO with SVM and applied it toa classroom response system

44 CSO+WNN Nanda proposed a hybrid system bycombining wavelet neural network (WNN) and CSO al-gorithm In their proposal the CSO algorithm was used totrain the weights of WNN in order to obtain the near-op-timal weights [41]

45 BCSO+ SVM Mohamadeen et al built a classificationmodel based on BCSO and SVM and then applied it in apower system +e use of BCSO was to optimize SVMparameters [42]

46 CCSO+ANN Wang et al proposed designing an ANNthat can handle randomness fuzziness and accumulativetime effect in time series concurrently In their work theCSO algorithm was used to optimize the network structureand learning parameters at the same time [43]

47 CSOPSO+ANN Chittineni et al used CSO and PSOalgorithms to train ANN and then applied their method onstock market prediction +eir comparison results showedthat CSO algorithm performed better than the PSO algo-rithm [44]

48 CS-FLANN Kumar et al combined the CSO algorithmwith functional link artificial neural network (FLANN) todevelop an evolutionary filter to remove Gaussian noise [45]

5 Applications of CSO

+is section presents the applications of CSO algorithmwhich are categorized into seven groups namely electricalengineering computer vision signal processing systemmanagement and combinatorial optimization wireless andWSN petroleum engineering and civil engineering Asummary of the purposes and results of these applications isprovided in Table 2

51 Electrical Engineering CSO algorithm has been exten-sively applied in the electrical engineering field Hwang et alapplied both CSO and PSO algorithms on an electricalpayment system in order to minimize electricity costs forcustomers Results indicated that CSO is more efficient andfaster than PSO in finding the global best solution [46]Economic load dispatch (ELD) and unit commitment (UC)are significant applications in which the goal is to reduce thetotal cost of fuel is a power system Hwang et al applied theCSO algorithm on economic load dispatch (ELD) of windand thermal generators [47] Faraji et al also proposedapplying binary cat swarm optimization (BCSO) algorithmon UC and obtained better results compared to the previousapproaches [48] UPFC stands for unified power flowcontroller which is an electrical device used in transmissionsystems to control both active and reactive power flowsKumar and Kalavathi used CSO algorithm to optimizeUPFC in order to improve the stability of the system [49]Lenin and Reddy also applied ADCSO on reactive powerdispatch problemwith the aim tominimize active power loss[50] Improving available transfer capability (ATC) is verysignificant in electrical engineering Nireekshana et al usedCSO algorithm to regulate the position and control pa-rameters of SVC and TCSC with the aim of maximizingpower transfer transactions during normal and contingencycases [51] +e function of the transformers is to deliverelectricity to consumers Determining how reliable thesetransformers are in a power system is essential Moha-madeen et al proposed a classification model to classify thetransformers according to their reliability status [42] +emodel was built based on BCSO incorporation with SVM+e results are then compared with a similar model based onBPSO It is shown that BCSO is more efficient in optimizingthe SVM parameters Wang et al proposed designing anANN that can handle randomness fuzziness and accu-mulative time effect in time series concurrently [43] In theirwork the CSO algorithm has been used to optimize thenetwork structure and learning parameters at the same time+en the model was applied to two applications which wereindividual household electric power consumption fore-casting and Alkaline-surfactant-polymer (ASP) flooding oilrecovery index forecasting in oilfield development +ecurrent source inverter (CSI) is a conventional kind of powerinverter topologies Hosseinnia and Farsadi combined


Table 2 +e purposes and results of using CSO algorithm in various applications

Purpose Results RefCSO applied on electrical payment system in order tominimize electricity cost for customers CSO outperformed PSO [46]

CSO applied on economic load dispatch (ELD) ofwind and thermal generator CSO outperformed PSO [47]

BCSO applied on unit commitment (UC) CSO outperformed LR ICGA BF MILP ICA andSFLA [48]

Applied CSO algorithm on UPFC to increase thestability of the system

IEEE 6-bus and 14-bus networks were used in thesimulation experiments and desirable results were

achieved[49]

Applied ADCSO on reactive power dispatch problemto minimize active power loss

IEEE 57-bus system was used in the simulationexperiments in which ADCSO outperformed 16

other optimization algorithms[50]

Applied CSO algorithm to regulate the position andcontrol parameters of SVC and TCSC to improveavailable transfer capability (ATC)

IEEE 14-bus and IEEE 24-bus systems were used inthe simulation experiments in which the system

provided better results after adopting CSO[51]

Building a classification model based on BCSO andSVM to classify the transformers according to theirreliability status

+e model performed better compared to a similarmodel which was based on BPSO and VSM [42]

Applied CSO to optimize the network structure andlearning parameters of an ANN model namedCPNN-CSO which is used to predict householdelectric power consumption

CPNN-CSO outperformed ANFIS and similarmethods with no CSO such as PNN and CPNN [43]

Applied CSO and selective harmonic elimination(SHE) algorithm on current source inverter (CSI)

CSO successfully optimized the switching parametersof CSI and hence minimized the total harmonic

distortion[52]

Applied both CSO PCSO PSO-CFA and ACO-ABCon distributed generation units on distributionnetworks

IEEE 33-bus and IEEE 69-bus distribution systemswere used in the simulation experiments and CSO

outperformed the other algorithms[53]

AppliedMCSO onMPPTto achieve global maximumpower point (GMPP) tracking

MCSO outperformed PSO MPSO DE GA and HCalgorithms [54]

Applied BCSO to optimize the location of phasormeasurement units and reduce the required numberof PMUs

IEEE 14-bus and IEEE 30-bus test systems were usedin the simulation BCSO outperformed BPSO

generalized integer linear programming and effectivedata structure-based algorithm

[55]

Used CSO algorithm to identify the parameters ofsingle and double diode models in solar cell system

CSO outperformed PSO GA SA PS Newton HSGGHS IGHS ABSO DE and LMSA [56]

Applied CSO and SVM to classify studentsrsquo facialexpression

+e results show 100 classification accuracy for theselected 9 face expressions [39]

Applied CSO and SVM to classify studentsrsquo facialexpression +e system achieved satisfactory results [40]

Applied CSO-GA-PSOSVM to classify studentsrsquofacial expression +e system achieved 99 classification accuracy [23]

Applied CSO HCSO and ICSO in block matching forefficient motion estimation

+e system reduced computational complexity andprovided faster convergence [16 17 57]

Used CSO algorithm to retrieve watermarks similarto the original copy

CSO outperformed PSO and PSO time-varyinginertia weight factor algorithms [58 59]

Sabah used EHCSO in an object-tracking system toobtain further efficiency and accuracy

+e system yielded desirable results in terms ofefficiency and accuracy [60]

Used BCSO as a band selection method forhyperspectral images BCSO outperformed PSO [61]

Used CSO and multilevel thresholding for imagesegmentation CSO outperformed PSO [62]

Used CSO and multilevel thresholding for imagesegmentation PSO outperformed CSO [63]

Used CSO ANN and wavelet entropy to build anAUD identification system CSO outperformed GA IGA PSO and CSPSO [64]

Used CSO and FLANN to remove the unwantedGaussian noises from CT images

+e proposed system outperformed mean filter andadaptive Wiener filter [45]

Used CSO with L-BFGS-B technique to registernonrigid multimodal images +e system yielded satisfactory results [65]

Used CSO in image enhancement to optimizeparameters of the histogram stretching technique PSO outperformed CSO [66]

Used CSO algorithm for IIR system identification CSO outperformed GA and PSO [67]


Table 2 Continued

Purpose Results RefApplied CSO to do direct and inverse modeling oflinear and nonlinear plants CSO outperformed GA and PSO [68]

Used CSO and SVM for electrocardiograms signalclassification

Optimizing SVM parameters using CSO improvedthe system in terms of accuracy [38]

Applied CSO to increase reliability in a task allocationsystem CSO outperformed GA and PSO [69 70]

Applied CSO on JSSP+e benchmark instances were taken from OR-

Library CSO yielded desirable results compared tothe best recorded results in the dataset reference

[71]

Applied BCSO on JSSP ACO outperformed CSO and cuckoo searchalgorithms [72]

Applied CSO on FSSPCarlier Heller and Reeves benchmark instances wereused CSO can solve problems of up to 50 jobs

accurately[73]

Applied CSO on OSSP CSO performs better than six metaheuristicalgorithms in the literature [74]

Applied CSO on JSSP CSO performs better than some conventionalalgorithms in terms of accuracy and speed [75]

Applied CSO on bag-of-tasks and workflowscheduling problems in cloud systems

CSO performs better than PSO and two otherheuristic algorithms [76]

Applied CSO on TSP and QAP

+e benchmark instances were taken from TSPLIBand QAPLIB +e results show that CSO

outperformed the best results recorded in thosedataset references

[77]

Comparison between CSO cuckoo search and bat-inspired algorithm to solve TSP problem

+e benchmark instances are taken from STPLIB+eresults show that CSO falls behind the other

algorithms[78]

Applied CSO and MCSO on workflow scheduling incloud systems CSO performs better than PSO [79]

Applied BCSO on workflow scheduling in cloudsystems BCSO performs better than PSO and BPSO [80]

Applied BCSO on SCP BCSO performs better than ABC [81]

Applied BCSO on SCP BCSO performs better than binary teaching-learning-based optimization (BTLBO) [82 83]

Used a CSO as a clustering mechanism in webservices CSO performs better than K-means [84]

Applied hybrid CSO-GA-SA to find the overlappingcommunity structures

Very good results were achieved Silhouettecoefficient was used to verify these results in which

was between 07 and 09[25]

Used CSO to optimize the network structures forpinning control CSO outperformed a number of heuristic methods [85]

Applied CSO with local search refining procedure toaddress high school timetabling problem

CSO outperformed genetic algorithm (GA)evolutionary algorithm (EA) simulated annealing(SA) particle swarm optimization (PSO) and

artificial fish swarm (AFS)

[24]

BCSO with dynamic mixture ratios to address themanufacturing cell design problem

BCSO can effectively tackle the MCDP problemregardless of the scale of the problem [86]

Used CSO to find the optimal reservoir operation inwater resource management CSO outperformed GA [87]

Applied CSO to classify the the feasibility of smallloans in banking systems

CSO resulted in 76 of accuracy in comparison to64 resulted from OLR procedure [88]

Used CSO AEM and RPT to build a groundwatermanagement systems

CSO outperformed a number of metaheuristicalgorithms in addressing groundwater management

problem[89]

Applied CSO to solve the multidocumentsummarization problem CSO outperformed harmonic search (HS) and PSO [90]

Used CSO and (RPCM) to address groundwaterresource management CSO outperformed a similar model based on PSO [91]

Applied CSO-CS to solve VRPTW

CSO-CS successfully solves the VRPTW problem+e results show that the algorithm convergencesfaster by increasing population and decreasing cdc

parameter

[32]


selective harmonic elimination (SHE) in corporation withCSO algorithm and then applied it on current source in-verter (CSI) [52] +e role of the CSO algorithm was tooptimize and tune the switching parameters and minimizetotal harmonic distortion El-Ela et al [53] used CSO andPCSO to find the optimal place and size of distributedgeneration units on distribution networks Guo et al [54]used MCSO algorithm to propose a novel maximum powerpoint tracking (MPPT) approach to obtain global maximumpower point (GMPP) tracking Srivastava et al used BCSOalgorithm to optimize the location of phasor measurementunits and reduce the required number of PMUs [55] Guoet al used CSO algorithm to identify the parameters of singleand double diode models in solar cell models [56]

52 Computer Vision Facial emotion recognition is a bio-metric approach to identify human emotion and classifythem accordingly Lin et al and Wang and Wu [39 40]proposed a classroom response system by combining theCSO algorithm with support vector machine to classifystudentrsquos facial expressions Vivek and Reddy also usedCSO-GA-PSOSVM algorithm for the same purpose [23]

Block matching in video processing is computationallyexpensive and time consuming Hadi and Sabah used CSOalgorithm in block matching for efficient motion estimation[57] +e aim was to decrease the number of positions thatneeds to be calculated within the search window during theblock matching process ie to enhance the performanceand reduce the number of iterations without the degradationof the image quality +e authors further improved theirwork and achieved better results by replacing the CSO al-gorithm with HCSO and ICSO in [16 17] respectivelyKalaiselvan et al and Lavanya and Natarajan [58 59] usedCSO Algorithm to retrieve watermarks similar to theoriginal copy In video processing object tracking is theprocess of determining the position of a moving object overtime using a camera Hadi and Sabah used EHCSO in anobject-tracking system for further enhancement in terms ofefficiency and accuracy [60] Yan et al used BCSO as a bandselection method for hyperspectral images [61] In computervision image segmentation refers to the process of dividingan image into multiple parts Ansar and Bhattacharya andKarakoyun et al [62 63] proposed using CSO algorithmincorporation with the concept of multilevel thresholdingfor image segmentation purposes Zhang et al combined

Table 2 Continued

Purpose Results RefApplied CSO and K-median to detect overlappingcommunity in social networks

CSO and K-median provides better modularity thansimilar models based on PSO and BAT algorithm [92]

Applied MOCSO fitness sharing and fuzzymechanism on CR design

MOCSO outperformed MOPSO NSGA-II andMOBFO [93 94]

Applied CSO and five other metaheuristic algorithmsto design a CR engine

CSO outperformed the GA PSO DE BFO and ABCalgorithms [95]

Applied EPCSO on WSN to be used as a routingalgorithm

EPCSO outperformed AODV a ladder diffusionusing ACO and a ladder diffusion using CSO [33]

Applied CSO on WSN in order to solve optimalpower allocation problem

PSO is marginally better for small networksHowever CSO outperformed PSO and cuckoo search

algorithm[96]

Applied CSO on WSN to optimize cluster headselection

+e proposed system outperformed the existingsystems by 75 [97]

Applied CSO on CR based smart gridcommunication network to optimize channelallocation

+e proposed system obtains desirable results forboth fairness-based and priority-based cases [98]

Applied CSO in WSN to detect optimal location ofsink nodes

CSO outperformed PSO in reducing total powerconsumption [99 100]

Applied CSO on time modulated concentric circularantenna array to minimize the sidelobe level ofantenna arrays and enhance the directivity

CSO outperformed RGA PSO and DE algorithms [101]

Applied CSO to optimize the radiation patterncontrolling parameters for linear antenna arrays

CSO successfully tunes the parameters and providesoptimal designs of linear antenna arrays [102]

Applied Cauchy mutated CSO to make linearaperiodic arrays where the goal was to reducesidelobe level and control the null positions

+e proposed system outperformed both CSO andPSO [103]

Applied CSO and analytical formula-based objectivefunction to optimize well placements CSO outperformed DE algorithm [104]

Applied CSO to optimize well placementsconsidering oilfield constraints during development CSO outperformed GA and DE algorithms [105]

CSO applied to optimize the network structure andlearning parameters of an ANN model which is usedto predict an ASP flooding oil recovery index

+e system successfully forecast the ASP flooding oilrecovery index [42]

Applied CSO to build an identification model todetect early cracks in beam type structures

CSO yields a desirable accuracy in detecting earlycracks [106]


wavelet entropy ANN and CSO algorithm to develop analcohol use disorder (AUD) identification system [64]Kumar et al combined the CSO algorithm with functionallink artificial neural network (FLANN) to remove the un-wanted Gaussian noises from CT images [45] Yang et alcombined CSO with L-BFGS-B technique to register non-rigid multimodal images [65] Ccedilam employed CSO algo-rithm to tune the parameters in the histogram stretchingtechnique for the purpose of image enhancement [66]

53 Signal Processing IIR filter stands for infinite impulseresponse It is a discrete-time filter which has applications insignal processing and communication Panda et al usedCSO algorithm for IIR system identification [67] +e au-thors also applied CSO algorithm as an optimizationmechanism to do direct and inverse modeling of linear andnonlinear plants [68] Al-Asadi combined CSO Algorithmwith SVM for electrocardiograms signal classification [38]

54 System Management and Combinatorial OptimizationIn parallel computing optimal task allocation is a keychallenge Shojaee et al [69 70] proposed using CSO al-gorithm to maximize system reliability +ere are three basicscheduling problems namely open shop job shop and flowshop +ese problems are classified as NP-hard and havemany real-world applications +ey coordinate assigningjobs to resources at particular times where the objective is tominimize time consumption However their difference ismainly in having ordering constraints on operationsBouzidi and Riffi applied the BCSO algorithm on jobscheduling problem (JSSP) in [71] +ey also made acomparative study between CSO and two other meta-heuristic algorithms namely cuckoo search (CS) algorithmand the ant colony optimization (ACO) for JSSP in [72]+en they used the CSO algorithm to solve flow shopscheduling (FSSP) [73] and open shop scheduling problems(OSSP) as well [74] Moreover Dani et al also applied CSOalgorithm on JSSP in which they used a nonconventionalapproach to represent cat positions [75] Maurya and Tri-pathi also applied CSO algorithm on bag-of-tasks andworkflow scheduling problems in cloud systems [76]Bouzidi and Riffi applied CSO algorithm on the travelingsalesman problem (TSP) and the quadratic assignmentproblem (QAP) which are two combinatorial optimizationproblems [77] Bouzidi et al also made a comparative studybetween CSO algorithm cuckoo search algorithm and bat-inspired algorithm for addressing TSP [78] In cloudcomputing minimizing the total execution cost while al-locating tasks to processing resources is a key problemBilgaiyan et al applied CSO and MCSO algorithms onworkflow scheduling in cloud systems [79] In additionKumar et al also applied BCSO on workflow scheduling incloud systems [80] Set cover problem (SCP) is considered asan NP-complete problem Crawford et al successfully ap-plied the BCSO Algorithm to this problem [81]+ey furtherimproved this work by using Binarization techniques andselecting different parameters for each test example sets[82 83] Web services provide a standardized

communication between applications over the web whichhave many important applications However discoveringappropriate web services for a given task is challengingKotekar and Kamath used a CSO-based approach as aclustering algorithm to group service documents accordingto their functionality similarities [84] Sarswat et al appliedHybrid CSO-GA-SA to detect the overlapping communitystructures and find the near-optimal disjoint communities[25] Optimizing the problem of controlling complex net-work systems is critical in many areas of science and en-gineering Orouskhani et al applied CSO algorithm toaddress a number of problems in optimal pinning con-trollability and thus optimized the network structure [85]Skoullis et al combined the CSO algorithm with local searchrefining procedure and applied it on high school timetablingproblem [24] Soto et al combined BCSO with dynamicmixture ratios to organize the cells in manufacturing celldesign problem [86] Bahrami et al applied a CSO algorithmon water resource management where the algorithm wasused to find the optimal reservoir operation [87] Kencanaet al used CSO algorithm to classify the feasibility of smallloans in banking systems [88] Majumder and Eldhocombined the CSO algorithm with the analytic elementmethod (AEM) and reverse particle tracking (RPT) to modelnovel groundwater management systems [89] Rautray andBalabantaray used CSO algorithm to solve the multidocu-ment summarization problem [90] +omas et al combinedradial point collocation meshfree (RPCM) approach withCSO algorithm to be used in the groundwater resourcemanagement [91] Pratiwi created a hybrid system bycombining the CSO algorithm and crow search (CS) algo-rithm and then used it to address the vehicle routingproblem with time windows (VRPTW) [32] Naem et alproposed a modularity-based system by combining the CSOalgorithm with K-median clustering technique to detectoverlapping community in social networks [92]

55 Wireless and WSN +e ever-growing wireless devicespush researchers to use electromagnetic spectrum bandsmore wisely Cognitive radio (CR) is an effective dynamicspectrum allocation in which spectrums are dynamicallyassigned based on a specific time or location Pradhan andPanda in [93 94] combined MOCSO with fitness sharingand fuzzy mechanism and applied it on CR design+ey alsoconducted a comparative analysis and proposed a gener-alized method to design a CR engine based on six evolu-tionary algorithms [95] Wireless sensor network (WSN)refers to a group of nodes (wireless sensors) that form anetwork to monitor physical or environmental conditions+e gathered data need to be forwarded among the nodesand each node requires having a routing path Kong et alproposed applying enhanced parallel cat swarm optimiza-tion (EPCSO) algorithm in this area as a routing algorithm[33] Another concern in the context of WSN is minimizingthe total power consumption while satisfying the perfor-mance criteria So Tsiflikiotis and Goudos addressed thisproblem which is known as optimal power allocationproblem and for that three metaheuristic algorithms were


presented and compared [96] Moreover Pushpalatha andKousalya applied CSO in WSN for optimizing cluster headselection which helps in energy saving and availablebandwidth [97] Alam et al also applied CSO algorithm in aclustering-based method to handle channel allocation (CA)issue between secondary users with respect to practicalconstraints in the smart grid environment [98]+e authorsof [99 100] used the CSO algorithm to find the optimallocation of sink nodes in WSN Ram et al applied CSOalgorithm to minimize the sidelobe level of antenna arraysand enhance the directivity [101] Ram et al used CSO tooptimize controlling parameters of linear antenna arraysand produce optimal designs [102] Pappula and Ghoshalso used Cauchy mutated CSO to make linear aperiodicarrays where the goal was to reduce sidelobe level andcontrol the null positions [103]

56 Petroleum Engineering CSO algorithm has also beenapplied in the petroleum engineering field For example itwas used as a good placement optimization approach byChen et al in [104 105] Furthermore Wang et al used CSOalgorithm as an ASP flooding oil recovery index forecastingapproach [43]

57 Civil Engineering Ghadim et al used CSO algorithm tocreate an identification model that detects early cracks inbuilding structures [106]

6 Performance Evaluation

Many variants and applications of CSO algorithm werediscussed in the above sections However benchmarkingthese versions and conducting a comparative analysis be-tween them were not feasible in this work +is is becausefirstly their source codes were not available Secondlydifferent test functions or datasets have been used duringtheir experiments In addition since the emergence of CSOalgorithm many novel and powerful metaheuristic algo-rithms have been introduced However the literature lacks acomparative study between CSO algorithm and these newalgorithms +erefore we conducted an experiment in

which the original CSO algorithm was compared againstthree new and robust algorithms which were dragonflyalgorithm (DA) [6] butterfly optimization algorithm (BOA)[7] and fitness dependent optimizer (FDO) [8] For this 23traditional and 10 modern benchmark functions were used(see Figure 3) which illustrates the general framework forconducting the performance evaluation process It is worthmentioning that for four test functions BOA returnedimaginary numbers and we set ldquoNArdquo for them

61 Traditional Benchmark Functions +is group includesthe unimodal and multimodal test functions Unimodal testfunctions contain one single optimumwhile multimodal testfunctions contain multiple local optima and usually a singleglobal optimum F1 to F7 are unimodal test functions(Table 3) which are employed to experiment with the globalsearch capability of the algorithms Furthermore F8 to F23are multimodal test functions which are employed to ex-periment with the local search capability of the algorithmsRefer to [107] for the detailed description of unimodal andmultimodal functions

62 Modern Benchmark Functions (CEC 2019) +ese set ofbenchmark functions also called composite benchmarkfunctions are complex and difficult to solve +e CEC01 toCEC10 functions as shown in Table 3 are of these typeswhich are shifted rotated expanded and combined versionsof traditional benchmark functions Refer to [108] for thedetailed description of modern benchmark functions

+e comparison results for CSO and other algorithmsare given in Table 3 in the form of mean and standarddeviations For each test function the algorithms are exe-cuted for 30 independent runs For each run 30 searchagents were searching over the course of 500 iterationsParameter settings are set as defaults for all algorithms andnothing was changed

It can be noticed fromTable 3 that the CSO algorithm is acompetitive algorithm for the modern ones and providesvery satisfactory results In order to perceive the overallperformance of the algorithms they are ranked as shown inTable 4 according to different benchmark function groups It

CSO and its competitive algorithms

Classical and modern benchmark functions

Ranking (Friedman test)

Wilcoxon matched-pairs signed-rank test

(confirm the results)

Default control parameters30 independent runs30 search agents500 iterations

(i)(ii)

(iii)(iv)

Figure 3 General framework of the performance evaluation process


Table 3 Comparison results of CSO algorithm with modern metaheuristic algorithms

CSO DA BOA FDOfminFunctions AV STD AV STD AV STD AV STD

F1 350E minus 14 634E minus 14 1524805 2378914 101E minus 11 166E minus 12 213E minus 23 106E minus 22 0F2 268E minus 08 261E minus 08 1458012 0869819 465E minus 09 463E minus 10 0047175 0188922 0F3 717E minus 09 116E minus 08 136259 1519406 108E minus 11 171E minus 12 239E minus 06 128E minus 05 0F4 0010352 0007956 3262584 2112636 525E minus 09 553E minus 10 493E minus 08 909E minus 08 0F5 8587858 0598892 3749048 6915889 8935518 002146 2158376 3966721 0F6 1151759 0431511 1207847 1797414 104685 0346543 715E minus 22 280E minus 21 0F7 0026026 0015039 0035679 0023538 0001513 000056 0612389 0299315 0F8 minus 285511 3591697 minus 281414 432944 NA NA minus 105021 1518877 minus 4189829times 5F9 2401772 6480946 2653478 1120011 286796 2017813 7940883 4110302 0F10 3754226 1680534 2827344 1042434 300E minus 09 116E minus 09 776E minus 15 246E minus 15 0F11 0355631 019145 0680359 0353454 135E minus 13 627E minus 14 0175694 0148586 0F12 1900773 1379549 2083215 1436402 0130733 0084891 7737715 4714534 0F13 1160662 053832 1072302 1327413 0451355 0138253 4724571 6448214 0F14 0998004 339E minus 07 1064272 0252193 152699 0841504 2448453 1766953 1F15 0001079 000117 0005567 0012211 0000427 987E minus 05 0001492 0003609 000030F16 minus 103162 153E minus 05 minus 103163 476E minus 07 NA NA minus 100442 0149011 minus 10316F17 0304253 181E minus 06 0304251 0 0310807 0004984 0397887 517E minus 15 0398F18 3003667 0004338 3000003 122E minus 05 3126995 0211554 3 237E minus 07 3F19 minus 38625 000063 minus 386262 000037 NA NA minus 386015 0003777 minus 386F20 minus 330564 0045254 minus 325226 0069341 NA NA minus 306154 0380813 minus 332F21 minus 988163 090859 minus 728362 2790655 minus 444409 0383552 minus 419074 2664305 minus 101532F22 minus 102995 0094999 minus 837454 2726577 minus 41496 0715469 minus 489633 3085016 minus 104028F23 minus 100356 1375583 minus 640669 2892797 minus 412367 0859409 minus 403276 2517357 minus 105363CEC01 158E+ 09 171E+ 09 38E+ 10 403E+ 10 5893069 1144572 4585278 2070763 1CEC02 1970367 0580672 8373248 1001326 1891597 0291311 4 328E minus 09 1CEC03 1370241 235E minus 06 1370263 0000673 1370321 0000617 137024 168E minus 11 1CEC04 1791984 5537322 3712471 4202062 209415 7707688 3308378 1681143 1CEC05 2671378 0171923 2571134 0304055 6176949 0708134 213924 0087218 1CEC06 1121251 0708359 1034469 1335367 1183069 0771166 1213326 0610499 1CEC07 3652358 164997 5343862 2400417 1043895 2153575 1204858 1382608 1CEC08 5499615 0484645 586374 051577 6337199 0359203 6102152 0769938 1CEC09 6325862 1295848 8501541 1690603 2270616 8114442 2 200E minus 10 1CEC10 2136829 006897 2129284 0176811 214936 0079492 2718282 452E minus 16 1

Table 4 Ranking of CSO algorithm compared to the modern metaheuristic algorithms

Test functions Ranking of CSO Ranking of DA Ranking of BOA Ranking of FDOF1 2 4 3 1F2 2 4 1 3F3 2 4 1 3F4 3 4 1 2F5 1 4 2 3F6 3 4 2 1F7 2 3 1 4F8 2 3 4 1F9 2 3 4 1F10 4 3 2 1F11 3 4 1 2F12 2 3 1 4F13 3 2 1 4F14 1 2 3 4F15 2 4 1 3F16 1 2 4 3F17 3 4 2 1F18 3 2 4 1F19 2 3 4 1F20 1 2 4 3F21 1 2 3 4


can be seen that CSO ranks first in the overall ranking andmultimodal test functions Additionally it ranks second inunimodal and CEC test functions (see Figure 4) +eseresults indicate the effectiveness and robustness of the CSOalgorithm +at being said these results need to be con-firmed statistically Table 5 presents the Wilcoxon matched-pairs signed-rank test for all test functions In more than85 of the results P value is less than 005 which provesthat the results are significant and we can reject the nullhypothesis that there is no difference between the means Itis worth mentioning that the performance of CSO can befurther evaluated by comparing it against other new algo-rithms such as donkey and smuggler optimization algorithm[109] modified grey wolf optimizer [110] BSA and its

variants [111] WOA and its variants [112] and othermodified versions of DA [113]

7 Conclusion and Future Directions

Cat swarm optimization (CSO) is a metaheuristic optimi-zation algorithm proposed originally by Chu et al [5] in2006 Henceforward many modified versions and applica-tions of it have been introduced However the literaturelacks a detailed survey in this regard +erefore this paperfirstly addressed this gap and presented a comprehensivereview including its developments and applications

CSO showed its ability in tackling different and complexproblems in various areas However just like any other

Table 4 Continued

Test functions Ranking of CSO Ranking of DA Ranking of BOA Ranking of FDOF22 1 2 4 3F23 1 2 3 4Cec01 3 4 2 1Cec02 3 4 2 1Cec03 2 3 4 1Cec04 2 3 4 1Cec05 3 2 4 1Cec06 2 1 3 4Cec07 2 3 4 1Cec08 1 2 4 3Cec09 2 3 4 1Cec10 3 2 4 1Total 70 97 91 72Overall ranking 2121212 2939394 2757576 2181818F1ndashF7 subtotal 15 27 11 17F1ndashF7 ranking 2142857 3857143 1571429 2428571F8ndashF23 subtotal 32 43 45 40F8ndashF23 ranking 2 26875 28125 25CEC01ndashCEC10 subtotal 23 27 35 15CEC01ndashCEC10 ranking 23 27 35 15

CSO DA BOA FDO

Overall rankingF1ndashF7 ranking

F8ndashF23 rankingCEC01ndashCEC10 ranking

0

05

1

15

2

25

3

35

4

Rank

ing

Figure 4 Ranking of algorithms according to different groups of test functions


metaheuristic algorithm CSO algorithm possesses strengthsand weaknesses +e tracing mode resembles the globalsearch process while the seeking mode resembles the localsearch process +is algorithm enjoys a significant propertyfor which these two modes are separated and independent+is enables researchers to easily modify or improve thesemodes and hence achieve a proper balance between ex-ploration and exploitation phases In addition fast con-vergence is another strong point of this algorithm whichmakes it a sensible choice for those applications that requirequick responses However the algorithm has a high chanceof falling into local optima known as premature conver-gence which can be considered as the main drawback of thealgorithm

Another concern was the fact that CSO algorithm wasnot given a chance to be compared against new algorithmssince it has been mostly measured up against PSO and GAalgorithms in the literature To address this a performanceevaluation was conducted to compare CSO against threenew and robust algorithms For this 23 traditional bench-mark functions and 10 modern benchmark functions wereused +e results showed the outperformance of CSO al-gorithm in which it ranked first in general +e significanceof these results was also confirmed by statistical methods

+is indicates that CSO is still a competitive algorithm in thefield

In the future the algorithm can be improved in manyaspects for example different techniques can be adapted tothe tracing mode in order to solve the premature conver-gence problem or transformingMR parameter is static in theoriginal version of CSO Transforming this parameter into adynamic parameter might improve the overall performanceof the algorithm

Conflicts of Interest

+e authors declare that there are no conflicts of interestregarding the publication of this paper

References

[1] X S Yang Nature-Inspired Metaheuristic AlgorithmsLuniver Press UK 2010

[2] S Kirkpatrick C D Gelatt and M P Vecchi ldquoOptimizationby simulated annealingrdquo Science vol 220 no 4598pp 671ndash680 1983

[3] J Kennedy ldquoParticle swarm optimizationrdquo Encyclopedia ofMachine Learning pp 760ndash766 Springer Berlin Germany2010

[4] S C Chu and P W Tsai ldquoComputational intelligence basedon the behavior of catsrdquo International Journal of InnovativeComputing Information and Control vol 3 no 1pp 163ndash173 2007

[5] S C Chu P W Tsai and J S Pan ldquoCat swarm optimi-zationrdquo in Proceedings of the Pacific Rim InternationalConference on Artificial Intelligence pp 854ndash858 SpringerGuilin China August 2006

[6] S Mirjalili ldquoDragonfly algorithm a new meta-heuristicoptimization technique for solving single-objective discreteand multi-objective problemsrdquo Neural Computing and Ap-plications vol 27 no 4 pp 1053ndash1073 2016

[7] S Arora and S Singh ldquoButterfly optimization algorithm anovel approach for global optimizationrdquo Soft Computingvol 23 no 3 pp 715ndash734 2019

[8] J M Abdullah and T Ahmed ldquoFitness dependent optimizerinspired by the bee swarming reproductive processrdquo IEEEAccess vol 7 pp 43473ndash43486 2019

[9] Y Sharafi M A Khanesar and M Teshnehlab ldquoDiscretebinary cat swarm optimization algorithmrdquo in Proceedings ofthe 2013 3rd International Conference on Computer Controlamp Communication (IC4) pp 1ndash6 IEEE Karachi PakistanSeptember 2013

[10] P M Pradhan and G Panda ldquoSolving multiobjectiveproblems using cat swarm optimizationrdquo Expert Systemswith Applications vol 39 no 3 pp 2956ndash2964 2012

[11] P W Tsai J S Pan S M Chen B Y Liao and S P HaoldquoParallel cat swarm optimizationrdquo in Proceedings of the 2008International Conference on Machine Learning and Cyber-netics vol 6 pp 3328ndash3333 IEEE Kunming China July2008

[12] B Santosa and M K Ningrum ldquoCat swarm optimization forclusteringrdquo in Proceedings of the 2009 International Con-ference of Soft Computing and Pattern Recognition pp 54ndash59IEEE December 2009

[13] P-W Tsai J-S Pan S-M Chen and B-Y Liao ldquoEnhancedparallel cat swarm optimization based on the Taguchi

Table 5 Wilcoxon matched-pairs signed-rank test

Test functions CSO vs DA CSO vs BOA CSO vs FDOF1 lt00001 lt00001 lt00001F2 lt00001 lt00001 00003F3 lt00001 lt00001 02286F4 lt00001 lt00001 lt00001F5 lt00001 00879 00732F6 00008 0271 lt00001F7 0077 lt00001 lt00001F8 0586 NA lt00001F9 02312 03818 lt00001F10 00105 lt00001 lt00001F11 lt00001 lt00001 00002F12 04 lt00001 lt00001F13 lt00001 lt00001 00185F14 04 lt00001 00003F15 00032 00004 09515F16 lt00001 NA lt00001F17 lt00001 lt00001 lt00001F18 lt00001 lt00001 lt00001F19 02109 NA 06554F20 00065 NA lt00001F21 00057 lt00001 lt00001F22 01716 lt00001 lt00001F23 lt00001 lt00001 lt00001cec01 lt00001 lt00001 lt00001cec02 0001 lt00001 lt00001cec03 00102 lt00001 lt00001cec04 00034 lt00001 lt00001cec05 01106 lt00001 lt00001cec06 00039 00007 lt00001cec07 00002 lt00001 lt00001cec08 00083 lt00001 lt00001cec09 0115 lt00001 lt00001cec10 00475 lt00001 lt00001


methodrdquo Expert Systems with Applications vol 39 no 7pp 6309ndash6319 2012

[14] M Orouskhani M Mansouri andM Teshnehlab ldquoAverage-inertia weighted cat swarm optimizationrdquo in Proceedings ofthe International Conference in Swarm Intelligencepp 321ndash328 Springer Chongqing China June 2011

[15] M Orouskhani Y Orouskhani M Mansouri andM Teshnehlab ldquoA novel cat swarm optimization algorithmfor unconstrained optimization problemsrdquo InternationalJournal of Information Technology and Computer Sciencevol 5 no 11 pp 32ndash41 2013

[16] I Hadi and M Sabah ldquoEnhanced hybrid cat swarm opti-mization based on fitness approximationmethod for efficientmotion estimationrdquo International Journal of Hybrid Infor-mation Technology vol 7 no 6 pp 345ndash364 2014

[17] I Hadi and M Sabah ldquoImprovement cat swarm optimiza-tion for efficient motion estimationrdquo International Journal ofHybrid Information Technology vol 8 no 1 pp 279ndash2942015

[18] Y Kumar and G Sahoo ldquoAn improved cat swarm opti-mization algorithm based on opposition-based learning andCauchy operator for clusteringrdquo Journal of InformationProcessing Systems vol 13 no 4 pp 1000ndash1013 2017

[19] X Nie W Wang and H Nie ldquoChaos quantum-behaved catswarm optimization algorithm and its application in the PVMPPTrdquo Computational Intelligence and Neurosciencevol 2017 Article ID 1583847 11 pages 2017

[20] N Kanwar N Gupta K R Niazi and A Swarnkar ldquoIm-proved cat swarm optimization for simultaneous allocationof DSTATCOM and DGs in distribution systemsrdquo Journal ofRenewable Energy vol 2015 Article ID 189080 10 pages2015

[21] Y Kumar and P K Singh ldquoImproved cat swarm optimi-zation algorithm for solving global optimization problemsand its application to clusteringrdquoApplied Intelligence vol 48no 9 pp 2681ndash2697 2017

[22] PW Tsai J S Pan P Shi and B Y Liao ldquoA new frameworkfor optimization based-on hybrid swarm intelligencerdquoHandbook of Swarm Intelligence pp 421ndash449 SpringerBerlin Germany 2011

[23] T V Vivek and G R Reddy ldquoA hybrid bioinspired algo-rithm for facial emotion recognition using CSO-GA-PSO-SVMrdquo in Proceedings of the 2015 Fifth International Con-ference on Communication Systems and Network Technolo-gies (CSNT) pp 472ndash477 IEEE April 2015

[24] V I Skoullis I X Tassopoulos and G N BeligiannisldquoSolving the high school timetabling problem using a hybridcat swarm optimization based algorithmrdquo Applied SoftComputing vol 52 pp 277ndash289 2017

[25] A Sarswat V Jami and R M Guddeti ldquoA novel two-stepapproach for overlapping community detection in socialnetworksrdquo Social Network Analysis and Mining vol 7 no 1p 47 2017

[26] K-C Lin Y-H Huang J C Hung and Y-T Lin ldquoFeatureselection and parameter optimization of support vectormachines based on modified cat swarm optimizationrdquo In-ternational Journal of Distributed Sensor Networks vol 11no 7 Article ID 365869 2015

[27] P Mohapatra S Chakravarty and P K Dash ldquoMicroarraymedical data classification using kernel ridge regression andmodified cat swarm optimization based gene selection sys-temrdquo Swarm and Evolutionary Computation vol 28pp 144ndash160 2016

[28] L Pappula and D Ghosh ldquoCat swarm optimization withnormal mutation for fast convergence of multimodalfunctionsrdquo Applied Soft Computing vol 66 pp 473ndash4912018

[29] K-C Lin K-Y Zhang Y-H Huang J C Hung and N YenldquoFeature selection based on an improved cat swarm opti-mization algorithm for big data classificationrdquoGe Journal ofSupercomputing vol 72 no 8 pp 3210ndash3221 2016

[30] M Zhao ldquoA novel compact cat swarm optimization based ondifferential methodrdquo Enterprise Information Systemsvol 2018 pp 1ndash25 2018

[31] H Siqueira E Figueiredo M Macedo C J SantanaC J Bastos-Filho and A A Gokhale ldquoBoolean binary catswarm optimization algorithmrdquo in Proceedings of the 2018IEEE Latin American Conference on Computational Intelli-gence (LA-CCI) pp 1ndash6 IEEE Guadalajara Mexico No-vember 2018

[32] A B Pratiwi ldquoA hybrid cat swarm optimization-crow searchalgorithm for vehicle routing problem with time windowsrdquoin Proceedings of the 2017 2nd International Conferences onInformation Technology Information Systems and ElectricalEngineering (ICITISEE) pp 364ndash368 IEEE YogyakartaIndonesia November 2017

[33] L Kong J-S Pan P-W Tsai S Vaclav and J-H Ho ldquoAbalanced power consumption algorithm based on enhancedparallel cat swarm optimization for wireless sensor networkrdquoInternational Journal of Distributed Sensor Networks vol 11no 3 Article ID 729680 2015

[34] Y Kumar and G Sahoo ldquoAn improved cat swarm opti-mization algorithm for clusteringrdquo Computational Intelli-gence in Data Mining-Volume 1 vol 2015 pp 187ndash197Springer New Delhi India

[35] A Baldominos Y Saez and P Isasi ldquoHybridizing evolu-tionary computation and deep neural networks an approachto handwriting recognition using committees and transferlearningrdquo Complexity vol 2019 Article ID 295230416 pages 2019

[36] J P T Yusiong ldquoOptimizing artificial neural networks usingcat swarm optimization algorithmrdquo International Journal ofIntelligent Systems and Applications vol 5 no 1 pp 69ndash802012

[37] M Orouskhani M Mansouri Y Orouskhani andM Teshnehlab ldquoA hybrid method of modified cat swarmoptimization and gradient descent algorithm for trainingANFISrdquo International Journal of Computational Intelligenceand Applications vol 12 no 2 Article ID 1350007 2013

[38] H A Al-Asadi ldquoNew hybrid (SVMs-CSOA) architecture forclassifying electrocardiograms signalsrdquo International Journalof Advanced Research in Artificial Intelligence (IJARAI)vol 4 no 5 2015

[39] K C Lin RW Lin S J Chen C R You and J L Chai ldquo+eclassroom response system based on affective computingrdquo inProceedings of the 2010 3rd IEEE International Conference onUbi-Media Computing pp 190ndash197 IEEE Jinhua ChinaJuly 2010

[40] W Wang and J Wu ldquoNotice of retraction emotion recog-nition based on CSOampSVM in e-learningrdquo in Proceedings ofthe Seventh International Conference on Natural Computa-tion vol 1 pp 566ndash570 IEEE Shanghai China July 2011

[41] S J Nanda ldquoAWNN-CSO model for accurate forecasting ofchaotic and nonlinear time seriesrdquo in Proceedings of the 2015IEEE International Conference on Signal Processing Infor-matics Communication and Energy Systems (SPICES)pp 1ndash5 IEEE Kozhikode India February 2015


[42] K I Mohamadeen R M Sharkawy and M M SalamaldquoBinary cat swarm optimization versus binary particle swarmoptimization for transformer health index determinationrdquo inProceedings of the 2014 International Conference on Engi-neering and Technology (ICET) pp 1ndash5 IEEE Cairo EgyptApril 2014

[43] B Wang S Xu X Yu and P Li ldquoTime series forecastingbased on cloud process neural networkrdquo InternationalJournal of Computational Intelligence Systems vol 8 no 5pp 992ndash1003 2015

[44] S Chittineni V Mounica K Abhilash S C Satapathy andP P Reddy ldquoA comparative study of CSO and PSO trainedartificial neural network for stock market predictionrdquo inProceedings of the International Conference on Computa-tional Science Engineering and Information Technologypp 186ndash195 Springer Tirunelveli India September 2011

[45] M Kumar S K Mishra and S S Sahu ldquoCat swarm opti-mization based functional link artificial neural network filterfor Gaussian noise removal from computed tomographyimagesrdquo Applied Computational Intelligence and SoftComputing vol 2016 Article ID 6304915 6 pages 2016

[46] J C Hwang J C Chen J S Pan and Y C Huang ldquoCSO andPSO to solve optimal contract capacity for high tensioncustomersrdquo in Proceedings of the 2009 International Con-ference on Power Electronics and Drive Systems (PEDS)pp 246ndash251 IEEE Taipei Taiwan November 2009

[47] J C Hwang J C Chen J S Pan and Y C Huang ldquoCSOalgorithm for economic dispatch decision of hybrid gener-ation systemrdquo in Proceedings of the 10th WSEAS Interna-tional Conference on Applied Informatics andCommunications and 3rd WSEAS International Conferenceon Biomedical Electronics and Biomedical Informaticspp 81ndash86 World Scientific and Engineering Academy andSociety (WSEAS) Taipei Taiwan August 2010

[48] I Faraji A Z Bargabadi and Z Hejrati Application ofBinary Cat Swarm Optimization Algorithm for Unit Com-mitment problem inGe First National Conference on Meta-Heuristic Algorithms and Geir Applications in Engineeringand Science Fereydunkenar Iran August 2014

[49] G N Kumar and M S Kalavathi ldquoDynamic load models forvoltage stability studies with a solution of UPFC using CSOrdquoInternational Journal of Computer Applications vol 116no 10 pp 27ndash32 2015

[50] K Lenin and B R Reddy ldquoReduction of active power loss byusing adaptive cat swarm optimizationrdquo Indonesian Journalof Electrical Engineering and Informatics (IJEEI) vol 2 no 3pp 111ndash118 2014

[51] T Nireekshana G Kesava Rao and S Sivanaga RajuldquoAvailable transfer capability enhancement with FACTSusing cat swarm optimizationrdquo Ain Shams EngineeringJournal vol 7 no 1 pp 159ndash167 2016

[52] H Hosseinnia and M Farsadi ldquoUtilization cat swarm op-timization algorithm for selected harmonic elemination incurrent source inverterrdquo International Journal of PowerElectronics and Drive Systems (IJPEDS) vol 6 no 4pp 888ndash896 2015

[53] A A El-Ela R A El-Sehiemy A M Kinawy and E S AlildquoOptimal placement and sizing of distributed generationunits using different cat swarm optimization algorithmsrdquo inProceedings of the 2016 Eighteenth International Middle EastPower Systems Conference (MEPCON) pp 975ndash981 IEEECairo Egypt December 2016

[54] L Guo Z Meng Y Sun and L Wang ldquoA modified catswarm optimization based maximum power point tracking

method for photovoltaic system under partially shadedconditionrdquo Energy vol 144 pp 501ndash514 2018

[55] A Srivastava and S Maheswarapu ldquoOptimal PMU place-ment for complete power system observability using binarycat swarm optimizationrdquo in Proceedings of the 2015 Inter-national Conference on Energy Economics and Environment(ICEEE) pp 1ndash6 IEEE Greater Noida India March 2015

[56] L Guo Z Meng Y Sun and L Wang ldquoParameter iden-tification and sensitivity analysis of solar cell models with catswarm optimization algorithmrdquo Energy Conversion andManagement vol 108 pp 520ndash528 2016

[57] I Hadi and M Sabah ldquoA novel block matching algorithmbased on cat swarm optimization for efficient motion esti-mationrdquo International Journal of Digital Content Technologyand Its Applications vol 8 no 6 p 33 2014

[58] G Kalaiselvan A Lavanya and V Natrajan ldquoEnhancing theperformance of watermarking based on cat swarm optimi-zation methodrdquo in Proceedings of the 2011 InternationalConference on Recent Trends in Information Technology(ICRTIT) pp 1081ndash1086 IEEE Chennai India June 2011

[59] A Lavanya and V Natarajan ldquoAnalyzing the performance ofwatermarking based on swarm optimization methodsrdquoAdvances in Computing and Information TechnologySpringer pp 167ndash176 Berlin Germany

[60] I Hadi and M Sabah ldquoAn enriched 3D trajectory generatedequations for the most common path of multiple objecttrackingrdquo International Journal of Multimedia and Ubiq-uitous Engineering vol 10 no 6 pp 53ndash76 2015

[61] L Yan X Yan-Qiu and W Li-Hai ldquoHyperspectral di-mensionality reduction of forest types based on cat swarmalgorithmrdquo Ge Open Automation and Control SystemsJournal vol 7 no 1 2015

[62] W Ansar and T Bhattacharya ldquoA new gray image seg-mentation algorithm using cat swarm optimizationrdquo inProceedings of the 2016 International Conference on Com-munication and Signal Processing (ICCSP) pp 1004ndash1008IEEE Chennai India April 2016

[63] M Karakoyun N A Baykan and M Hacibeyoglu ldquoMulti-level thresholding for image segmentation with swarm op-timization algorithmsrdquo International Research Journal ofElectronics amp Computer Engineering vol 3 no 3 p 1 2017

[64] Y D Zhang Y Sui J Sun G Zhao and P Qian ldquoCat swarmoptimization applied to alcohol use disorder identificationrdquoMultimedia Tools and Applications vol 77 no 17pp 22875ndash22896 2018

[65] F Yang M Ding X Zhang W Hou and C Zhong ldquoNon-rigid multi-modal medical image registration by combiningL-BFGS-B with cat swarm optimizationrdquo Information Sci-ences vol 316 pp 440ndash456 2015

[66] H B Ccedilam S Akccedilakoca A Elbir H O Ilhan and N Aydınldquo+e performance evaluation of the cat and particle swarmoptimization techniques in the image enhancementrdquo inProceedings of the 2018 Electric Electronics Computer ScienceBiomedical Engineeringsrsquo Meeting (EBBT) pp 1ndash4 IEEEIstanbul Turkey April 2018

[67] G Panda P M Pradhan and B Majhi ldquoIIR system iden-tification using cat swarm optimizationrdquo Expert Systems withApplications vol 38 no 10 pp 12671ndash12683 2011

[68] G Panda P M Pradhan and B Majhi ldquoDirect and inversemodeling of plants using cat swarm optimizationrdquo Hand-book of Swarm Intelligence pp 469ndash485 Springer BerlinGermany 2011

[69] R Shojaee H R Faragardi S Alaee and N Yazdani ldquoA newcat swarm optimization-based algorithm for reliability-


oriented task allocation in distributed systemsrdquo in Pro-ceedings of the 6th International Symposium on Telecom-munications (IST) pp 861ndash866 IEEE Tehran IranNovember 2012

[70] R Shojaee H R Faragardi and N Yazdani ldquoFrom reliabledistributed system toward reliable cloud by cat swarm op-timizationrdquo International Journal of Information andCommunication vol 5 no 4 pp 9ndash18 2013

[71] A Bouzidi and M E Riffi ldquoCat swarm optimization to solvejob shop scheduling problemrdquo in Proceedings of the 2014Gird IEEE International Colloquium in Information Scienceand Technology (CIST) pp 202ndash205 IEEE Tetuan MoroccoOctober 2014

[72] A Bouzidi and M E Riffi ldquoA comparative study of threepopulation-based metaheuristics for solving the JSSPrdquo inEurope and MENA Cooperation Advances in Informationand Communication Technologies pp 235ndash243 SpringerCham Switzerland 2017

[73] A Bouzidi and M E Riffi ldquoCat swarm optimization to solveflow shop scheduling problemrdquo Journal of Georetical ampApplied Information Technology vol 72 no 2 2015

[74] A Bouzidi M E Riffi and M Barkatou ldquoCat swarm op-timization for solving the open shop scheduling problemrdquoJournal of Industrial Engineering International vol 15 no 2pp 367ndash378 2019

[75] V Dani A Sarswat V Swaroop S Domanal andR M Guddeti ldquoFast convergence to near optimal solutionfor job shop scheduling using cat swarm optimizationrdquo inProceedings of the International Conference on PatternRecognition and Machine Intelligence pp 282ndash288 SpringerKolkata India December 2017

[76] A K Maurya and A K Tripathi ldquoDeadline-constrainedalgorithms for scheduling of bag-of-tasks and workflows incloud computing environmentsrdquo in Proceedings of the 2ndInternational Conference on High Performance CompilationComputing and Communications pp 6ndash10 ACM HongKong China March 2018

[77] A Bouzidi and M E Riffi ldquoDiscrete cat swarm optimizationalgorithm applied to combinatorial optimization problemsrdquoin Proceedings of the 2014 5th Workshop on Codes Cryp-tography and Communication Systems (WCCCS) pp 30ndash34IEEE El Jadida Morocco November 2014

[78] S Bouzidi M E Riffi and A Bouzidi ldquoComparative analysisof three metaheuristics for solving the travelling salesmanproblemrdquo Transactions on Machine Learning and ArtificialIntelligence vol 5 no 4 2017

[79] S Bilgaiyan S Sagnika and M Das ldquoWorkflow schedulingin cloud computing environment using cat swarm optimi-zationrdquo in Proceedings of the 2014 IEEE International Ad-vance Computing Conference (IACC) pp 680ndash685 IEEEGurgaon India February 2014

[80] B Kumar M Kalra and P Singh ldquoDiscrete binary catswarm optimization for scheduling workflow applications incloud systemsrdquo in Proceedings of the 2017 3rd InternationalConference on Computational Intelligence amp Communica-tion Technology (CICT) pp 1ndash6 IEEE Ghaziabad IndiaFebruary 2017

[81] B Crawford R Soto N Berrıos et al ldquoA binary cat swarmoptimization algorithm for the non-unicost set coveringproblemrdquo Mathematical Problems in Engineering vol 2015Article ID 578541 8 pages 2015

[82] B Crawford R Soto N Berrios and E Olguin ldquoSolving theset covering problem using the binary cat swarm optimi-zation metaheuristic World academy of science engineering

and technologyrdquo International Journal of MathematicalComputational Physical Electrical and Computer Engi-neering vol 10 no 3 pp 104ndash108 2016

[83] B Crawford R Soto N Berrios and E Olguın ldquoCat swarmoptimization with different binarization methods forsolving set covering problemsrdquo in Artificial IntelligencePerspectives in Intelligent Systems pp 511ndash524 SpringerCham Switzerland 2016

[84] S Kotekar and S S Kamath ldquoEnhancing service discoveryusing cat swarm optimisation based web service clusteringrdquoPerspectives in Science vol 8 pp 715ndash717 2016

[85] Y Orouskhani M Jalili and X Yu ldquoOptimizing dynamicalnetwork structure for pinning controlrdquo Scientific Reportsvol 6 no 1 Article ID 24252 2016

[86] R Soto B Crawford A Aste Toledo C Castro F Paredesand R Olivares ldquoSolving the manufacturing cell designproblem through binary cat swarm optimization with dy-namic mixture ratiosrdquo Computational Intelligence andNeuroscience vol 2019 Article ID 4787856 16 pages 2019

[87] M Bahrami O Bozorg-Haddad and X Chu ldquoApplication ofcat swarm optimization algorithm for optimal reservoiroperationrdquo Journal of Irrigation and Drainage Engineeringvol 144 no 1 Article ID 04017057 2017

[88] E N Kencana N Kiswanti and K Sari ldquo+e application ofcat swarm optimisation algorithm in classifying small loanperformancerdquo Journal of Physics Conference Series vol 893Article ID 012037 2017

[89] P Majumder and T I Eldho ldquoA new groundwater man-agement model by coupling analytic element method andreverse particle tracking with cat swarm optimizationrdquoWater Resources Management vol 30 no 6 pp 1953ndash19722016

[90] R Rautray and R C Balabantaray ldquoCat swarm optimizationbased evolutionary framework for multi document sum-marizationrdquo Physica A Statistical Mechanics and its Appli-cations vol 477 pp 174ndash186 2017

[91] A +omas P Majumdar T I Eldho and A K RastogildquoSimulation optimization model for aquifer parameter es-timation using coupled meshfree point collocation methodand cat swarm optimizationrdquo Engineering Analysis withBoundary Elements vol 91 pp 60ndash72 2018

[92] A A Naem L M El Bakrawy and N I Ghali ldquoA hybrid catoptimization and K-median for solving community detec-tionrdquo Asian Journal of Applied Sciences vol 5 no 5 2017

[93] P M Pradhan and G Panda ldquoPareto optimization ofcognitive radio parameters usingmultiobjective evolutionaryalgorithms and fuzzy decision makingrdquo Swarm and Evolu-tionary Computation vol 7 pp 7ndash20 2012

[94] P M Pradhan and G Panda ldquoCooperative spectrum sensingin cognitive radio network using multiobjective evolutionaryalgorithms and fuzzy decision makingrdquo Ad Hoc Networksvol 11 no 3 pp 1022ndash1036 2013

[95] P M Pradhan and G Panda ldquoComparative performanceanalysis of evolutionary algorithm based parameter opti-mization in cognitive radio engine a surveyrdquo Ad HocNetworks vol 17 pp 129ndash146 2014

[96] A Tsiflikiotis and S K Goudos ldquoOptimal power allocationin wireless sensor networks using emerging nature-inspiredalgorithmsrdquo in Proceedings of the 2016 5th InternationalConference on Modern Circuits and Systems Technologies(MOCAST) pp 1ndash4 IEEE +essaloniki Greece May 2016

[97] A Pushpalatha and G Kousalya ldquoA prolonged network lifetime and reliable data transmission aware optimal sink


relocation mechanismrdquo Cluster Computing vol 22 no S5pp 12049ndash12058 2018

[98] S Alam A N Malik I M Qureshi S A Ghauri andM Sarfraz ldquoClustering-based channel allocation scheme forneighborhood area network in a cognitive radio based smartgrid communicationrdquo IEEE Access vol 6 pp 25773ndash257842018

[99] V Snasel L Kong P Tsai and J S Pan ldquoSink nodeplacement strategies based on cat swarm optimization al-gorithmrdquo Journal of Network Intelligence vol 1 no 2pp 52ndash60 2016

[100] P W Tsai L Kong S Vaclav J S Pan V Istanda andZ Y Hu ldquoUtilizing cat swarm optimization in allocating thesink node in the wireless sensor network environmentrdquo inProceedings of the 2016 Gird International Conference onComputing Measurement Control and Sensor Network(CMCSN) pp 166ndash169 IEEE Matsue Japan May 2016

[101] G Ram D Mandal R Kar and S P Ghoshal ldquoCat swarmoptimization as applied to time-modulated concentric cir-cular antenna array analysis and comparison with otherstochastic optimization methodsrdquo IEEE Transactions onAntennas and Propagation vol 63 no 9 pp 4180ndash41832015

[102] G Ram D Mandal S P Ghoshal and R Kar ldquoOptimalarray factor radiation pattern synthesis for linear antennaarray using cat swarm optimization validation by an elec-tromagnetic simulatorrdquo Frontiers of Information Technologyamp Electronic Engineering vol 18 no 4 pp 570ndash577 2017

[103] L Pappula and D Ghosh ldquoSynthesis of linear aperiodic arrayusing cauchy mutated cat swarm optimizationrdquo AEU-ndashInternational Journal of Electronics and Communicationsvol 72 pp 52ndash64 2017

[104] H Chen Q Feng X Zhang SWangW Zhou and Y GengldquoWell placement optimization using an analytical formula-based objective function and cat swarm optimization algo-rithmrdquo Journal of Petroleum Science and Engineeringvol 157 pp 1067ndash1083 2017

[105] C Hongwei F Qihong Z Xianmin W Sen Z Wenshengand L Fan ldquoWell placement optimization with cat swarmoptimization algorithm under oil field development con-straintsrdquo Journal of Energy Resources Technology vol 141no 1 Article ID 012902 2019

[106] R Hassannejad S Tasoujian and M R Alipour ldquoBreathingcrack identification in beam-type structures using cat swarmoptimization algorithmrdquo Modares Mechanical Engineeringvol 15 no 12 pp 17ndash24 2016

[107] S Mirjalili and A Lewis ldquo+e whale optimization algo-rithmrdquo Advances in Engineering Software vol 95 pp 51ndash672016

[108] K V Price N H Awad M Z Ali and P N SuganthanGe100-Digit Challenge Problem Definitions and EvaluationCriteria for the 100-Digit Challenge Special Session andCompetition on Single Objective Numerical OptimizationNanyang Technological University Singapore 2018

[109] A S Shamsaldin T A Rashid R A Agha N K Al-Salihiand M Mohammadi ldquoDonkey and smuggler optimizationalgorithm a collaborative working approach to path find-ingrdquo Journal of Computational Design and Engineeringvol 6 no 4 pp 562ndash583 2019

[110] T A Rashid D K Abbas and Y K Turel ldquoA multi hiddenrecurrent neural network with a modified grey wolf opti-mizerrdquo PLoS One vol 14 no 3 Article ID e0213237 2019

[111] B A Hassan and T A Rashid ldquoOperational framework forrecent advances in backtracking search optimisation

algorithm a systematic review and performance evaluationrdquoApplied Mathematics and Computation vol 370 Article ID124919 2019

[112] H M Mohammed S U Umar and T A Rashid ldquoA sys-tematic and meta-analysis survey of whale optimizationalgorithmrdquo Computational Intelligence and Neurosciencevol 2019 Article ID 8718571 25 pages 2019

[113] C M Rahman and T A Rashid ldquoDragonfly algorithm andits applications in applied science surveyrdquo ComputationalIntelligence and Neuroscience vol 2019 Article ID 929361721 pages 2019


Computer Games Technology

International Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom

Journal ofEngineeringVolume 2018

Advances in

FuzzySystems


Volume 2018


ReconfigurableComputing



Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

thinspArtificial Intelligence

Hindawiwwwhindawicom Volumethinsp2018


Civil EngineeringAdvances in


Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications


Hindawi

wwwhindawicom Volume 2018

Advances in

Multimedia


Biomedical Imaging



Engineering Mathematics


RoboticsJournal of



Computational Intelligence and Neuroscience


Mathematical Problems in Engineering

Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018


Human-ComputerInteraction

Advances in


Scientic Programming

Submit your manuscripts atwwwhindawicom

literature lacks a recent and detailed review of this algorithmIn addition since 2006 CSO has not been compared againstnovel algorithms ie it has been mostly compared with PSOalgorithm while many new algorithms have been introducedsince then So a question which arises is whether CSOcompetes with the novel algorithms or not +ereforeexperimenting CSO on a wider range of test functions andcomparing it with new and robust algorithms will furtherreveal the potential of the algorithm As a result the aims ofthis paper are as follows firstly provide a comprehensiveand detailed review of the state of art of CSO algorithm (seeFigure 1) which shows the general framework for con-ducting the survey secondly evaluate the performance ofCSO algorithm against modern metaheuristic algorithms+ese should hugely help researchers to further work in thedomain in terms of developments and applications

+e rest of the paper is organized as follows Section 2presents the original algorithm and its mathematicalmodeling Section 3 is dedicated to reviewing all modifiedversions and variants of CSO Section 4 summarizes thehybridizing CSO algorithm with ANN and other non-metaheuristic methods Section 5 presents applications ofthe algorithm and groups them according to their disci-plinary Section 6 provides performance evaluation whereCSO is compared against dragonfly algorithm (DA) [6]butterfly optimization algorithm (BOA) [7] and fitnessdependent optimizer (FDO) [8] Finally Section 7 providesthe conclusion and future directions

2 Original Cat Swarm Optimization Algorithm

+e original cat swarm optimization is a continuous andsingle-objective algorithm [4 5] It is inspired by resting andtracing behaviours of cats Cats seem to be lazy and spendmost of their time resting However during their rests theirconsciousness is very high and they are very aware of what ishappening around them So they are constantly observingthe surroundings intelligently and deliberately and whenthey see a target they start moving towards it quickly+erefore CSO algorithm is modeled based on combiningthese two main deportments of cats

CSO algorithm is composed of two modes namelytracing and seeking modes Each cat represents a solutionset which has its own position a fitness value and a flag+eposition is made up of M dimensions in the search spaceand each dimension has its own velocity the fitness valuedepicts how well the solution set (cat) is finally the flag is toclassify the cats into either seeking or tracing mode+us weshould first specify how many cats should be engaged in theiteration and run them through the algorithm +e best catin each iteration is saved into memory and the one at thefinal iteration will represent the final solution

21 General Structure of the Algorithms +e algorithm takesthe following steps in order to search for optimal solutions

(1) Specify the upper and lower bounds for the solutionsets

(2) Randomly generate N cats (solution sets) and spreadthem in the M dimensional space in which each cathas a random velocity value not larger than a pre-defined maximum velocity value

(3) Randomly classify the cats into seeking and tracingmodes according to MR MR is a mixture ratiowhich is chosen in the interval of [0 1] So forexample if a number of cats N is equal to 10 and MRis set to 02 then 8 cats will be randomly chosen to gothrough seeking mode and the other 2 cats will gothrough tracing mode

(4) Evaluate the fitness value of all the cats according tothe domain-specified fitness function Next the bestcat is chosen and saved into memory

(5) +e cats thenmove to either seeking or tracingmode(6) After the cats go through seeking or tracing mode

for the next iteration randomly redistribute the catsinto seeking or tracing modes based on MR

(7) Check the termination condition if satisfied termi-nate the program otherwise repeat Step 4 to Step 6

22 Seeking Mode +is mode imitates the resting behaviorof cats where four fundamental parameters play importantroles seeking memory pool (SMP) seeking range of theselected dimension (SRD) counts of dimension to change(CDC) and self-position considering (SPC) +ese valuesare all tuned and defined by the user through a trial-and-error method

SMP specifies the size of seeking memory for cats ie itdefines number of candidate positions in which one of themis going to be chosen by the cat to go to for example if SMPwas set to 5 then for each and every cat 5 new randompositions will be generated and one of them will be selectedto be the next position of the cat How to randomize the newpositions will depend on the other two parameters that areCDC and SRD CDC defines how many dimensions to bemodified which is in the interval of [0 1] For example if thesearch space has 5 dimensions and CDC is set to 02 then foreach cat four random dimensions out of the five need to bemodified and the other one stays the same SRD is themutative ratio for the selected dimensions ie it defines theamount of mutation and modifications for those dimensionsthat were selected by CDC Finally SPC is a Boolean valuewhich specifies whether the current position of a cat will

CSO and its variants with artificial neural

networks

Original CSO algorithm Variants of CSO

Applications of CSO

Figure 1 General framework for conducting the survey









Pi FSi minus FSb

11138681113868111386811138681113868111386811138681113868








Xkd Xkd + Vkd (4)


3 Variants of CSO









Start

Generate N cats





YesNoSeeking mode



copies of the cat




Tracing mode

Velocity valuegt

maximum value


No


Updatepositions


Terminate

Stop







Better [9]




[11]




H6 PSO-DVM











complex conditions

Better [19]






Better [21]








BBCSO

















































achieved[49]













distortion[52]









[55]























Table 2 Continued







[71]



accurately[73]








[77]



algorithms[78]













[24]








problem[89]





parameter

[32]





Table 2 Continued











algorithm[96]










































(i)(ii)

(iii)(iv)














Table 4 Continued


CSO DA BOA FDO



0

05

1

15

2

25

3

35

4

Rank

ing









References




































































































































Advances in

FuzzySystems


Volume 2018













Journal of

Journal of



Hindawi


Advances in

Multimedia


Biomedical Imaging





RoboticsJournal of









Volume 2018



Advances in











Pi FSi minus FSb

11138681113868111386811138681113868111386811138681113868








Xkd Xkd + Vkd (4)


3 Variants of CSO









Start

Generate N cats





YesNoSeeking mode



copies of the cat




Tracing mode

Velocity valuegt

maximum value


No


Updatepositions


Terminate

Stop







Better [9]




[11]




H6 PSO-DVM











complex conditions

Better [19]






Better [21]








BBCSO

















































achieved[49]













distortion[52]









[55]























Table 2 Continued







[71]



accurately[73]








[77]



algorithms[78]













[24]








problem[89]





parameter

[32]





Table 2 Continued











algorithm[96]










































(i)(ii)

(iii)(iv)














Table 4 Continued


CSO DA BOA FDO



0

05

1

15

2

25

3

35

4

Rank

ing









References




































































































































Advances in

FuzzySystems


Volume 2018













Journal of

Journal of



Hindawi


Advances in

Multimedia


Biomedical Imaging





RoboticsJournal of









Volume 2018



Advances in






Start

Generate N cats





YesNoSeeking mode



copies of the cat




Tracing mode

Velocity valuegt

maximum value


No


Updatepositions


Terminate

Stop







Better [9]




[11]




H6 PSO-DVM











complex conditions

Better [19]






Better [21]








BBCSO

















































achieved[49]













distortion[52]









[55]























Table 2 Continued







[71]



accurately[73]








[77]



algorithms[78]













[24]








problem[89]





parameter

[32]





Table 2 Continued











algorithm[96]










































(i)(ii)

(iii)(iv)














Table 4 Continued


CSO DA BOA FDO



0

05

1

15

2

25

3

35

4

Rank

ing









References




































































































































Advances in

FuzzySystems


Volume 2018













Journal of

Journal of



Hindawi


Advances in

Multimedia


Biomedical Imaging





RoboticsJournal of









Volume 2018



Advances in








Better [9]




[11]




H6 PSO-DVM











complex conditions

Better [19]






Better [21]








BBCSO

















































achieved[49]













distortion[52]









[55]























Table 2 Continued







[71]



accurately[73]








[77]



algorithms[78]













[24]








problem[89]





parameter

[32]





Table 2 Continued











algorithm[96]










































(i)(ii)

(iii)(iv)














Table 4 Continued


CSO DA BOA FDO



0

05

1

15

2

25

3

35

4

Rank

ing









References




































































































































Advances in

FuzzySystems


Volume 2018













Journal of

Journal of



Hindawi


Advances in

Multimedia


Biomedical Imaging





RoboticsJournal of









Volume 2018



Advances in















































achieved[49]













distortion[52]









[55]























Table 2 Continued







[71]



accurately[73]








[77]



algorithms[78]













[24]








problem[89]





parameter

[32]





Table 2 Continued











algorithm[96]










































(i)(ii)

(iii)(iv)














Table 4 Continued


CSO DA BOA FDO



0

05

1

15

2

25

3

35

4

Rank

ing









References




































































































































Advances in

FuzzySystems


Volume 2018













Journal of

Journal of



Hindawi


Advances in

Multimedia


Biomedical Imaging





RoboticsJournal of









Volume 2018



Advances in




Table 2 Continued







[71]



accurately[73]








[77]



algorithms[78]













[24]








problem[89]





parameter

[32]





Table 2 Continued











algorithm[96]










































(i)(ii)

(iii)(iv)














Table 4 Continued


CSO DA BOA FDO



0

05

1

15

2

25

3

35

4

Rank

ing









References




































































































































Advances in

FuzzySystems


Volume 2018













Journal of

Journal of



Hindawi


Advances in

Multimedia


Biomedical Imaging





RoboticsJournal of









Volume 2018



Advances in







Table 2 Continued











algorithm[96]










































(i)(ii)

(iii)(iv)














Table 4 Continued


CSO DA BOA FDO



0

05

1

15

2

25

3

35

4

Rank

ing









References




































































































































Advances in

FuzzySystems


Volume 2018













Journal of

Journal of



Hindawi


Advances in

Multimedia


Biomedical Imaging





RoboticsJournal of









Volume 2018



Advances in


























(i)(ii)

(iii)(iv)














Table 4 Continued


CSO DA BOA FDO



0

05

1

15

2

25

3

35

4

Rank

ing









References




































































































































Advances in

FuzzySystems


Volume 2018













Journal of

Journal of



Hindawi


Advances in

Multimedia


Biomedical Imaging





RoboticsJournal of









Volume 2018



Advances in















Table 4 Continued


CSO DA BOA FDO



0

05

1

15

2

25

3

35

4

Rank

ing









References




































































































































Advances in

FuzzySystems


Volume 2018













Journal of

Journal of



Hindawi


Advances in

Multimedia


Biomedical Imaging





RoboticsJournal of









Volume 2018



Advances in










References




































































































































Advances in

FuzzySystems


Volume 2018













Journal of

Journal of



Hindawi


Advances in

Multimedia


Biomedical Imaging





RoboticsJournal of









Volume 2018



Advances in























































































































Advances in

FuzzySystems


Volume 2018













Journal of

Journal of



Hindawi


Advances in

Multimedia


Biomedical Imaging





RoboticsJournal of









Volume 2018



Advances in




catswarmoptimizationalgorithm:asurveyand …downloads.hindawi.com/journals/cin/2020/4854895.pdf ·...

Documents