You are on page 1of 8

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

net/publication/343332379

Comparative Analysis of Ant Colony Optimization and Genetic Algorithm on


Solving Symmetrical Travelling Salesman Problem

Article  in  Journal of Advanced Research in Dynamical and Control Systems · July 2020


DOI: 10.5373/JARDCS/V12SP7/20202399

CITATION READS

1 422

3 authors:

Ghaisan Latheef Hao Zhan Gan


Asia Pacific University of Technology and Innovation Asia Pacific University of Technology and Innovation
3 PUBLICATIONS   1 CITATION    2 PUBLICATIONS   1 CITATION   

SEE PROFILE SEE PROFILE

Zailan Arabee Abdul Salam


Asia Pacific University of Technology and Innovation
5 PUBLICATIONS   26 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Ghaisan Latheef on 13 August 2020.

The user has requested enhancement of the downloaded file.


Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 07-Special Issue, 2020

Comparative Analysis of Ant Colony


Optimization and Genetic Algorithm on
Solving Symmetrical Travelling Salesman
Problem
Mohamed Ghaisan Latheef, Asia Pacific University of Technology & Innovation, Bukit Jalil, Kuala Lumpur, Malaysia.
E-mail: TP047492@mail.apu.edu.my
Gan Hao Zhan, Asia Pacific University of Technology & Innovation, Bukit Jalil, Kuala Lumpur, Malaysia.
E-mail: TP042328@mail.apu.edu.my
Zailan Arabee Bin Abdul Salam, Asia Pacific University of Technology & Innovation, Bukit Jalil, Kuala Lumpur, Malaysia.
E-mail: zailan@apu.edu.my
Rajermani Thinakaran, Asia Pacific University of Technology & Innovation, Bukit Jalil, Kuala Lumpur, Malaysia.
E-mail: rajermani@apu.edu.my
Abstract--- An implementation of Ant Colony Optimization(ACO) in Matlab is used for optimal travel route in
specific instances of the Symmetric Travelling Salesman Problem and compare the results with that of Genetic
Algorithm(GA) by running them on the same symmetric TSP models. Several papers on similar topics and
experiments conducted are reviewed. Statistical analysis methods are then used to attempt to find the model-specific
ideal values for pheromone and heuristic exponential weight, as well as the pheromone evaporation rate, parameters
not shared by GA, while keeping parameters shared by both algorithms the same. Even so, results conclude that GA
performs better on all metrics as compared to ACO – faster with less computational time per iteration proportional
to number of agents, while providing better solutions at all ranges. While ACO provides better solutions with fewer
iterations, it takes more time per iteration as compared to GA, which ends up providing better solutions in the same
timespan with more iterations. The possibility is mentioned that these results may be skewed as ACO has more
parameters and require more fine-tuning as compared to GA per model, which may not have been done perfectly.
There is also the possibility that the ACO implementation in MATLAB was coded in such a way that it added
unnecessary computational time. ACO is also an umbrella term for many variants unlike GA, such as the initial Ant
System which was later improved into Ant Colony System, so the implementation might be of an older ACO
variant.
Keywords--- Ant Colony Optimization, Combinatorial Optimization, Genetic Algorithm, Metaheuristics, Shortest
Path Problem, Travelling Salesman Problem.

I. Introduction
1.1 Travelling Salesman Problem
The Travelling Salesman Problem (TSP) is one of the most well-known non-deterministic polynomial-time hard
problems classified under Combinatorial Optimization, which means there is no exact algorithm specifically
designed to solve that problem in polynomial time. So, metaheuristic algorithms are used, which are problem-
independent algorithms that do not take into account the specificity of the problem unlike regular heuristic
algorithms, in order to solve the problem, or rather, as it often is, to try and achieve the closest to optimal solution as
the algorithm used possibly can.
The TSP is described as the following: There are cities and given distances between them. A travelling salesman
has to visit all of them, without going back and visiting another, and as the salesman wishes to spend the least time
possible on the journey, the route to take in order to minimize the total distance travelled that would still fulfil the
requirements above, must therefore be found.
TSP was first formulated as a mathematical problem in 1930 and is one of the most intensively studied problems
in optimization [1]. It is used as a benchmark for many optimization methods and algorithms as well as for practical
applications especially in transportation and logistics, such as finding the shortest route in navigation apps and
planning bus or train lines, but also in the areas that are entirely unrelated to routing, such as industrial robot control,

DOI: 10.5373/JARDCS/V12SP7/20202399
ISSN 1943-023X 2629
Received: 25 May 2020/Accepted: 25 June 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 07-Special Issue, 2020

starlight interferometer satellite positioning, chain diagram optimization, drilling holes on electrical circuit boards,
designing the layout of computer motherboard components, applications in crystallography, and many more.
1.2 Ant Colony Optimization
Ant Colony Optimization (ACO) is a metaheuristic technique initially proposed by Dorigo et al. as the Ant
System (AS) [1], and since then, several improvement on the original AS have been devised, such as the Ant Colony
System (ACS) [2], Rank-Based Ant System [3], and Min-Max Ant System (MMAS) [4]. The algorithms were also
later compared for performance by Dorigo [5] and using TSP for benchmarking in later works.
The ACO algorithm is inspired by real ant colonies and the way they function. The underlying idea was to use
several agents that simulate real ants. Ants’ behavior are governed by the goal of survival of the colony/community
and the propagation of the species rather than being focused on individualistic survival, working much like organs of
a larger organism. The behaviour that inspired ACO in particular however, is their foraging behaviour – how ants
can find shortest paths between food sources and their nest. When searching for food, ants initially explore the area
surrounding their nest in a random manner. While moving, ants leave a chemical pheromone trail on the ground,
which other ants can smell.
As soon as an ant finds a food source, it evaluates the food and carries some of it back to the nest. During the
return trip, the quantity of pheromone that an ant leaves on the ground depend on the evaluation the ant made of the
food. The pheromone trails will guide other ants to the food source, and if there are multiple sources, the food
sources evaluated as more or better quality would have more pheromone, and more ants would congregate there.
The indirect communication between the ants via pheromone trails enables them to find shortest paths between their
nest and food sources without actually having a centralized intelligence network.
ACO simulates an ant colony with artificial ants and pheromone in order to search for solutions. The parameters
the user can customize include number of iterations (i), number of agents/ants (m), pheromone exponential weight
(α), heuristic exponential weight (β), amount of pheromone (q) and pheromone evaporation rate (ρ). Different value
combinations of α, β, q, and ρ are suitable for different models, and must be adjusted depending on the problem for
the best degree of optimization achievable by ACO.
1.3 Genetic Algorithm
Genetic Algorithm (GA) is an optimization algorithm inspired by the mechanism of natural selection and natural
genetics, developed by John Holland, his colleagues and students at University of Michigan [6]. The string structure
exploit historical information will assign the search into better performance area within the search space, although
they locate randomly. Only the stronger string structure (species with randomize yet structured information) will
pass down their gene, the unfit species will face extinction. For every generation, a new set of species (strings) will
be created using the old bits and pieces from the fittest species, hence passing down their gene and information.
The offspring will inherit all the information from the parent and will be better than them, which the process
continues until the final fittest individual appears. There are 5 stages in GA, which are initial population, fitness
function, selection, crossover and mutation [7]. A population is the set of species when the process begins, each
individual is equivalent to a solution to the proposed problem. The parameters known as genes are used to
characterize the individual which merge with the binary to form a solution (chromosome).
In the algorithm, there is a function to values the fitness of an individual, the probability for the individual to be
chosen for reproduce is based on the score. After scoring, 2 parent chromosomes will be chosen to reproduce, where
they merge their data to produce an offspring, this process is known as crossover. In the crossover phase, a crossover
point will be selected randomly within the chromosome, parents will exchange their gene until they reached the
point, which an offspring is produced. Lastly mutation is certain of random bits in the offspring will be flipped to
maintain the diversity and evolution, but it will only occur with low random probability.
After all the phases complete and there are no more offspring produced, it means it is converged and the search
is complete. This is known as termination and the solutions are already provided to the problem at hand.

II. Materials and Methods


2.1 Hardware Configuration
For the purposes of this experiment, two PC configurations were used to run the algorithm on different TSP
models. A PC with Intel i7-8750H Hexa-Core CPU @ 2.20Ghz and 8GB RAM was used to run Berlin52, Eil51 and
Eil76. To run Oliver32, Qatar194 and Att532, a PC with Intel i7-6500U Dual-Core CPU @ 2.50Ghz was used.

DOI: 10.5373/JARDCS/V12SP7/20202399
ISSN 1943-023X 2630
Received: 25 May 2020/Accepted: 25 June 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 07-Special Issue, 2020

However, an Intel i3 CPU with 2 GB RAM minimum should be able to run the same experiments, albeit perhaps
slower.
2.2 Code and Software Required
The ACO and GA codes used are available on the MATLAB Central File Exchange, by Yarpiz [8] and Joseph
Kirk [9] respectively. Minimal changes to the MATLAB code were made. For ACO, it was consolidating all the
functions into a single file and pre-allocating memory to arrays with variable size in loops to slightly optimize the
code. For GA, the plot graph design was changed to reflect that of ACO, and vertical concatenation of x and y arrays
and transposing the matrix done so that dataset input for ACO and GA would be in identical format. However, as
most of the changes were mainly aesthetic and how it takes in input, all of the above should have no changes on how
the code implements the algorithm and processes the model (other than modified ACO code being marginally faster
in computation as compared to the original). It should be possible to recreate the experiment with the original codes
provided online.
To run the code, only MATLAB is needed, without any additional Toolboxes or Simulink. MATLAB version
R2018B was used for these results, but it has been run on version R2010A for backwards compatibility testing with
no issues.
2.3 TSP Datasets
The datasets used are all available online by searching the names, which are Oliver30, Att532, Berlin52, Eil51
and Eil76.
2.4 Previous Studies
In the work submitted by Ivan Brezina Jr. and Zuzana Čičková [10], they use the ACO to solve TSP. They
investigate the relation between the number of ants and the algorithm solution accuracy relative to the optimal
solution. They found that, while ceteris paribus (all other factors remaining constant) holds true, there is a
proportional positive link between the number of ants and the accuracy of the solution provided by the algorithm.
The article by Zaid Ameen Abduljabbar, Mustafa S. Khalefa, and Marzanah A. Jabar [11], also did comparison
on ACO and GA using randomly generated TSP datasets and found that ACO performed better in terms of number
of iterations taken and solution provided. These are similar conclusions to the paper by Nitin Chandekar and M.
Jayachandran Pillai in 2017 [12].
The paper by Khushboo and Mamta Arora [13] also supports the above conclusion that ACO is better for TSP
than GA, but provides execution time as a metric for judging efficiency as compared to iteration count unlike the
former two papers.
Another study by Mohammed Alhanjouri and Belal Alfarra [14] aimed to use both ACO and GA on TSP
datasets and directly compared the pair of solutions provided by the two algorithms for each dataset, in order to find
which is more suitable for TSP. With ACO, the parameters are largely dependent on the problem to be optimized,
the authors decided to try and determine the best parameter values though trial and error, then repeat for different
datasets.
Alhanjouri and Alfarra [14] also found that for ACO, it was far more difficult to select the best parameters as
there were less ideal values for a model, and more ideal combinations of value per model. For example, a slight
change in α alone might produce a significantly worse result but adjusting β as well might end up with a better than
initial solution. Thus, an overarching pattern was hard to identify, but they estimated general patterns where if p was
0, the algorithm did not converge, but if it was greater than or equal to 0.9, it converged to suboptimal solutions for
problems with more nodes/cities. However, disregarding the fact that they may not have found the optimal
parameters for ACO per model, based on the current data available to them, they postulated that GA is better than
ACO for solving TSP.
In GA, crossover probability determines how much the chromosomes of the parents are mixed, so that the
offspring end up being a mix of their parents or identical copies of one. Mutation probability determines how much
mutation there is, from no change to complete randomness. Population size is proportional to the search space
explored, but if increased too much beyond the number of cities, would only have negligible returns for
exponentially increased computational cost and time taken.

DOI: 10.5373/JARDCS/V12SP7/20202399
ISSN 1943-023X 2631
Received: 25 May 2020/Accepted: 25 June 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 07-Special Issue, 2020

By changing the GA parameter values, Alhanjouri and Alfarra found that the ideal Chromosome Population was
usually the number of cities multiplied by a factor of 10 and Crossover Probability as around 0.75 to 0.8 in most
models. However, Mutation Probability had to be fine adjusted for best performance as in case of 25 or more cities,
the results became very sensitive to any variation, where differences of 0.001 resulted in extremely different
solutions.
For ACO, the approaches to achieve the optimal solution are changing the parameters, the amounts of ants (m),
pheromone factor (α), heuristic factor (β), the residues coefficient of pheromone (ρ) and the amount of pheromone
(q), these parameters are the main factor that would affect the efficiency of the algorithm
Y. Feng concludes in her thesis that the best model of ACO from her experiments would be the parameters of
m=50 α=1 β=3 ρ=0.3 q=100 [15]. Each parameter is closely related to each other, but the major parameter
influencing the algorithm performance is the pheromone factor α, heuristic factor β and the residues coefficient of
pheromone ρ.
For general problem, ant colony algorithm is a good feedback mechanism, but still has some flaws even though
it has a lot of advantages. The algorithm gets the results very similar to the optimal result from the database, it would
be a sustainable algorithm if the parameter is fine-tuned based on the problem instances.
GA gives excellent result with small to medium size problem sets, while ACO has better performance with
larger instance, but it is likely to be trapped in its own local minima, which will be solved if given more time for
pheromone to evaporate [16].
Haroun, Jamal and Hicham [16] draws a conclusion that GA is fast, easy to implement and cost efficient, while
ACO is a greedier approach but provides better and quality results in large problems. GA is suitable for rapidly
obtaining optimal solution, although ACO consumes higher computational resources, but can also easily hybrid with
other algorithms. Both algorithms need to fine-tune the model to adapt with the problem in order to obtain better
results as they are sensitive to the parameter.

III. Results and Discussion


In this study, the authors compared the efficiency and effectiveness of ACO and GA for Combinatorial
Optimization and Shortest Path Problems. As such, the TSP model was used as it is highly regarded as somewhat an
industry standard in testing similar algorithms.
In order to provide a basis for direct side-by-side comparison, ACO and GA were used on the same well known
TSP datasets – all of which are available publicly online in the TSPLIB library – recording the results in Table 1 and
2: the name of the Dataset, the parameters used, the shortest distance achieved (Best Cost), the shortest actual
possible distance of the model (Optimal), and the time taken in seconds by the algorithm to reach the shortest
distance achieved (Time Elapsed (s)).
For all but Att532, which is by far the most complex in terms of number of nodes and thus computationally
expensive, GA performed the best in terms of both Best Cost and Time Elapsed metrics, and even for Att532’s
solution, took less time as compared to ACO, although providing a relatively subpar solution.
A particularly noteworthy fact is that GA was also able to find the optimal solution for Oliver32, a 32-node
dataset, in approximately 5 seconds, whereas ACO took 17.5 seconds to reach its shortest, not even optimal, after
which it stayed constant, stuck in local minima.
While choosing the parameters for ACO and GA, random parameter value combinations were blind tested and
those that gave better results for the particular model chosen. Through this, there was a noticeable trend such that
there were consistently better solutions produced by increasing the value of m, which corroborates the theory
postulated by Ivan Brezina Jr. and Zuzana Čičková [10] in their paper, even if the gains were minimal after m
reached a certain ratio as compared to the number of cities.
While their theory holds true, they did not document the increase of computational time due to increasing the
value of m. The authors’ tests conclude that there is a near exponential increase of time taken when the number of
ants are increased, which is a vital factor to take into account when running any algorithms, as they should be
effective, but also efficient when it comes to time. However, this may not be the fault of the algorithm, and instead
its specific MATLAB implementation used for this experiment.

DOI: 10.5373/JARDCS/V12SP7/20202399
ISSN 1943-023X 2632
Received: 25 May 2020/Accepted: 25 June 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 07-Special Issue, 2020

Table 1: ACO Results


Dataset i m α β ρ Best cost Optimal Time Elapsed (s)
Oliver32 100 200 1 3 0.3 425.6490 423.7406 17.517
Att532 200 1000 1 20 0.05 97860 86729 1688.3
Qatar174 200 200 1 3 0.3 10601.1563 9352 448.598
Berlin52 200 60 1 3 0.3 7548.9927 7542 24.391
Eil51 200 60 1 3 0.3 442.03 426 23.717
Eil76 200 80 1 3 0.3 569.7481 529 39.340
Table 2: GA Results
Dataset i m P(Mut.) P(Cross.) Best cost Optimal Time Elapsed (s)
Oliver32 200 30 Const. Const. 423.7406 423.7406 5.161
Att532 500 50000 Const. Const. 109218 86729 1437
Qatar174 1e5 200 Const. Const. 9778.5769 9352 423.879
Berlin52 1e4 52 Const. Const. 7544.3659 7542 7.602
Eil51 200 51 Const. Const. 429.4841 426 8.314
Eil76 1e4 76 Const. Const. 551.0075 529 12.296
The results of this paper also supports the conclusions reached by Mohammed Alhanjouri and Belal Alfarra [14],
where they compared both algorithms as well, and found GA to be superior for the particular task at hand – that of
Combinatorial Optimization problems.
Although, the authors would like to assert that there is a lack of data available in their article to make that
statement, coincidental veracity notwithstanding, as all the datasets used in the experiment had a lack of variation
when it came to number of cities, being merely 14 to 25 cities overall. These are fairly small when it comes to TSP,
as there are problems with hundreds and even thousands of cities. It is possible to speculate with only the above
information to support, that ACO may hypothetically be better with more cities than GA. In comparison, the
experiment outlined in this paper and conclusions made therein are from datasets as small as 32 to as large as 532
cities. These results partially agree with the claims made in the paper by Abduljabbar, Khalefa, and Jabar, [11] and
that of Chandekar and Pilla, [12] that while the results presented in this paper contradict their conclusion, ACO did
indeed take fewer iterations to produce similar results as compared to GA. However, something that both their
papers did not take into account or document, similar to Brezina and Čičková’s paper [10], is a lack comparison of
computational speed overall as well as per iteration without simply looking at number of iterations at face value, as
we found ACO took much longer per iteration compared to GA, even if the number of iterations were fewer.
The paper by Arora et. al [13] provides execution time of the experiments done on the datasets by ACO and GA,
however their results contradict those below, and they do not provide number of iterations.
A point to note is that the former three papers did not give the values for the parameters used when they tested
ACO and GA, making it hard to reproduce the experiments and verify any claims made.

IV. Limitations
The vastly more time taken by ACO could be explained by several factors, several of which could mean the
conclusion that GA performs better is erroneous, on account of flawed results.
Firstly, ACO has more parameters and require more fine-tuning as compared to GA per model, which may not
have been done perfectly. However, this is due to the nature of the algorithm. Metaheuristic algorithms should be
able to find approximately good solutions to all problems, however in ACO, not fine tuning the parameters
depending on the model results in far worse solutions. GA on the other hand, the values of Mutation and Crossover
Probabilities were kept constant for all TSP datasets and they still provided better solutions than ACO, which only
supports the conclusion made in this paper.
There is also the possibility that the ACO implementation in MATLAB was coded in such a way that it added
unnecessary computational time as compared to the MATLAB GA implementation used for the experiments. ACO
is also an umbrella term to many variants unlike GA, such as the initial Ant System (AS) which later on was
improved into Ant Colony System (ACS), so the implementation might be of an older ACO variant.

DOI: 10.5373/JARDCS/V12SP7/20202399
ISSN 1943-023X 2633
Received: 25 May 2020/Accepted: 25 June 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 07-Special Issue, 2020

V. Conclusion
While ACO provides better solutions with fewer iterations, it takes more time per iteration as compared to GA,
which ends up providing better solutions in the same timespan with more iterations. ACO in general provides better
performance across the board with datasets containing more cities. ACO has similar performance with GA with the
same iteration count i and value of m (ant/chromosome count), however it is infeasible to run ACO with the same
values as GA for i and m, as ACO is simply slower to compute. GA also has fewer dataset-dependent parameters,
making it easier and faster to run as compared to ACO, which you initially have to find the optimal parameters for
that dataset in order to ensure best performance that can compete with GA.
However, this does not mean ACO is necessarily a worse algorithm, as the "No Free Lunch" (NFL) theorem by
David Wolpert and William Macready postulates, ―that any two optimization algorithms are equivalent when their
performance is averaged across all possible problems" [17]. It is simply that ACO is relatively ill-suited for the
particular problem this paper focuses on.

Acknowledgements
The authors would also like to thank Yarpiz and Joseph Kirk, for providing the online redistributable MATLAB
codes of ACO and GA that was used for the experiments, saving much time and energy.

References
[1] D.L. Applegate, R.E. Bixby, V. Chvátal and W.J. Cook, The Traveling Salesman Problem: A
Computational Study, Princeton: Princeton University Press, 2007.
[2] M. Dorigo, V. Maniezzo and A. Colorni, "The Ant System: Optimization by a colony of cooperating
agents," IEEE Transactions on Systems, Man, and Cybernetics - Part B, vol. 26, no. 1, pp. 1-13, 1996.
[3] M. Dorigo and L.M. Gambardella, "Ant colony system: a cooperative learning approach to the traveling
salesman problem," IEEE Trans. Evolutionary Computation, vol. 1, no. 1, pp. 53-66, 1997.
[4] B. Bullnheimer, R. F. Hartl and C. Strauss, "A New Rank Based Version of the Ant System - A
Computational Study," Working Paper, vol. 1, no. 1, 1997.
[5] T. Stützle and H. H. Hoos, "MAX–MIN Ant System," Future Generation Computer Systems, vol. 16, no. 8,
pp. 889-914, 2000.
[6] M. Dorigo and T. Stützle, "ACO Algorithms for the Traveling Salesman Problem," Evolutionary
Algorithms in Engineering, pp. 163-183, 1999.
[7] D.E. Goldberg, Genetic Algorithm in Search, Optimization & Machine Learning, ADDISON-WESLEY
PUBLISHING COMPANY, INC, 1953.
[8] V. Mallawaarachchi, "Introduction to Genetic Algorithms — Including Example Code," 2017. [Online].
Available: https://towardsdatascience.com/introduction-to-genetic-algorithms-including-example-code-
e396e98d8bf3. [Accessed 2 December 2018].
[9] Yarpiz, "Ant Colony Optimization (ACO) - File Exchange - MATLAB Central," 2015. [Online]. Available:
https://www.mathworks.com/matlabcentral/fileexchange/52859-ant-colony-optimization-aco. [Accessed 12
October 2018].
[10] J. Kirk, "Traveling Salesman Problem - Genetic Algorithm - File Exchange - MATLAB Central," 2014.
[Online]. Available: https://www.mathworks.com/matlabcentral/fileexchange/13680-traveling-salesman-
problem-genetic-algorithm. [Accessed 4 November 2018].
[11] I.J. Brezina and Z. Čičková, "Solving the Travelling Salesman Problem Using the Ant Colony
Optimization," Management Information Systems, vol. 6, no. 4, pp. 10-14, 2011.
[12] Z.A. Abduljabbar, M.S. Khalefa and M.A. Jabar, "Comparison Between Ant Colony and Genetic
Algorithm Using Traveling Salesman Problem," International Journal of Soft Computing, vol. 8, no. 3, pp.
171-174, 2013.
[13] N. Chandekar and M. J. Pillai, "A Comparative Study of GA and ACO for Solving Travelling Salesman
Problem," International Journal of Mechanical And Production Engineering, vol. 5, no. 11, pp. 34-37,
2017.
[14] K. Arora and M. Arora, "Better Result for Solving TSP: GA versus ACO," International Journal of
Advanced Research in Computer Science and Software Engineering, vol. 6, no. 3, 2016.
[15] M. Alhanjouri and B. Alfarra, "Ant Colony versus Genetic Algorithm based on Travelling Salesman
Problem," International Journal of Computer Technology and Applications, vol. 2, no. 3, pp. 570-578,
2011.

DOI: 10.5373/JARDCS/V12SP7/20202399
ISSN 1943-023X 2634
Received: 25 May 2020/Accepted: 25 June 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 07-Special Issue, 2020

[16] Y. Feng, "Ant colony for the TSP - Thesis," 7 June 2010. [Online]. Available: https://www.diva-
portal.org/smash/get/diva2:518891/FULLTEXT01.pdf. [Accessed 5 December 2018].
[17] S.A. Haroun, B. Jamal and E.H. Hicham, "A Performance Comparison of GA and ACO Applied to,"
International Journal of Computer Applications (0975 – 8887), vol. 117, 2015.
[18] D.H. Wolpert and W.G. Macready, "No Free Lunch Theorems for Optimization," IEEE Transactions on
Evolutionary Computation, vol. 1, no. 1, pp. 67-82, 1997.

DOI: 10.5373/JARDCS/V12SP7/20202399
ISSN 1943-023X 2635
Received: 25 May 2020/Accepted: 25 June 2020

View publication stats

You might also like