Boosting GA Performance: A Fuzzy Approach to Uncertainty Issues
Involving Parameters in Genetic Algorithms
Jo
˜
ao Victor Ribeiro Ferro
1 a
, Jos
´
e Rubens da Silva Brito
1 b
, Rob
´
erio Jos
´
e Rog
´
erio dos Santos
2 c
,
Roberta Vilhena Vieira Lopes
1
and Evandro de Barros Costa
1
1
Computing Institute, Federal University of Alagoas, Av. Lourival Melo Mota, Maceio, Brazil
2
Eixo das Tecnologias, Campus do Sert
˜
ao, Federal University of Alagoas, Delmiro Gouveia, Brazil
Keywords:
Fuzzy Logic, Genetic Algorithms, Uncertainty, Optimization.
Abstract:
This article addresses issues involving two sources of uncertainty in the stochastic search problem based on
a genetic algorithm approach. We improve the mutation rate parameter by fuzzifying the population diver-
sity and the individual adaptation value. A relevant aspect of this investment is related to the fact that this
parameter, which presents uncertainty of the possibilistic type, directly interferes with the uncertainty of the
probabilistic type of the genetic algorithm and also in the convergence and quality of the solution found by
the genetic algorithm. Moreover, in parallel, we improve the understanding behavior of selection and replace-
ment methods. Experiments were carried out on the case study with the classic OneMax problem to evaluate
the performance of the proposed solution, analyzing aspects such as the convergence time, the quality of the
solution, and the diversity of the population. The results obtained through the treatment of uncertainty and
its impacts are presented in this article, showing relevant performance for the proposed algorithm, with the
respective treatment of uncertainties.
1 INTRODUCTION
One of the problems faced by stochastic optimiza-
tion models based on genetic algorithms (GA) is that
the speed of solution convergence is quite sensitive
to the choice of parameters such as mutation rate and
crossover rate. In addition, the process of defining
these parameters is also associated with the uncertain-
ties of which values to choose to improve the perfor-
mance of the algorithmic solution. For this reason,
genetic algorithms, by definition, have two sources
of uncertainty: inherent uncertainty in the adopted
stochastic model and the uncertainty of choosing the
best parameters for a good quality of the stochastic
search performance process.
In this sense, uncertainty is present in the GA
through the genetic operators of crossover, selection
of individuals, mutation, and the definition of the
stop state since the choice of parameterization of each
method is a process done in a non-automatic way and
based only on the programmer’s experience levels,
a
https://orcid.org/0000-0001-5806-2798
b
https://orcid.org/0000-0002-2291-0668
c
https://orcid.org/0000-0001-6304-0083
seen as a decision-making agent. In addition, we have
uncertainty linked to the very definition of the genetic
algorithm because it is a non-deterministic solution;
that is, each interaction can bring a different solution
but close to the optimal global solution.
In this way, we observe the need to reduce uncer-
tainty in the definition and selection of its key param-
eters because, on the one hand, the genetic mutation
operator conducts Exploration to identify yet unex-
plored solution subspaces i.e., those subspaces that
visit entirely new regions of a search space. On the
other hand, Exploitation is performed by crossover,
which operates on neighboring solution subspaces,
i.e., within the vicinity of previously visited points, to
find optimal local solutions (
ˇ
Crepin
ˇ
sek et al., 2013).
When doing an Exploration, genetic algorithms
employ the mutation rate, responsible for determining
the number of new search subspaces to be considered
for construction in the next iteration of the algorithm.
However, determining the optimal value for the muta-
tion rate is a challenge, given that different problems
may require different values.
Programmers determine the mutation rate choice,
which can be used as a default value by trial and error.
However, these ways of choosing the mutation rate
750
Ferro, J., Brito, J., Santos, R., Lopes, R. and Costa, E.
Boosting GA Performance: A Fuzzy Approach to Uncertainty Issues Involving Parameters in Genetic Algorithms.
DOI: 10.5220/0012389200003636
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 16th International Conference on Agents and Artificial Intelligence (ICAART 2024) - Volume 3, pages 750-757
ISBN: 978-989-758-680-4; ISSN: 2184-433X
Proceedings Copyright © 2024 by SCITEPRESS Science and Technology Publications, Lda.
may generate or not an efficient Exploration search
due to all sources of uncertainty in the decision pro-
cess of selecting the proper parameters.
In this article, we investigate the impact of uncer-
tainty sources of a genetic algorithm in the combi-
nation of selection and substitution methods on the
GA convergence time and population diversity, eval-
uating the behavior of two types of uncertainty: the
uncertainty inherent to the stochasticity of the algo-
rithm itself (probabilistic uncertainty) and the possi-
bilistic uncertainty present in the definition of popu-
lation diversity, as well as in the definition of chro-
mosomal fitness as the main parameters, due to the
inherent vagueness of these concepts, in the view of
the decision-making agent (the programmer of the ge-
netic algorithm).
Thus, our proposal focused on the development
of a self-adaptive genetic algorithm, investing in the
mutation rate with variation in the combination of se-
lection and substitution methods, but also in the use
of fuzzy logic to deal with the conceptual vagueness
present in the definition’s parameters of population di-
versity and chromosomal fitness.
In summary, the main contribution of this article
is to expose how uncertainty affects the genetic al-
gorithm due to its stochastic nature, but also how to
take advantage of the correct choice of evolutionary
methods responsible for the search for the optimal in-
dividual and the convergence time of the genetic al-
gorithm, dealing directly with conceptual vagueness,
with fuzzy sets, associated with population variability
and chromosomal aptitude of the best-adapted indi-
viduals.
2 SOURCES OF UNCERTAINTY
IN THE GENETIC
ALGORITHM
In the evaluation of a genetic algorithm, we find two
sources of uncertainty:
1) uncertainty inherent to the stochastic GA model
2) uncertainty inherent in random choices of param-
eterization by the programmer
In GA, uncertainty is based on these two points.
On the one hand, we have to model through the
genetic operators of mutation, crossover, algorithm
stopping point, and method of choosing the most
adapted individuals from one generation to another;
on the other hand, we have the programmer (decision-
making agent) with his prior knowledge on how to
properly define these parameters, which often need
to be modified and tested during application develop-
ment, being adjusted empirically and often based on
vague definitions about quality of the parameters to
choose.
The GA itself uses the uncertainty of the prob-
abilistic type in the process of search and self-
adaptation of individuals better adapted to the solu-
tion environment, being carried out through distribu-
tions of random variables that underlie the process of
stochastic search and directly influence the conver-
gence itself, still being related to GA parameteriza-
tion.
On the other hand, the possibilistic uncertainty is
due to the vagueness in the precision of the informa-
tion regarding the choice of parameters that best fit
a stochastic search problem and is associated with the
programmer’s decision (and his degree of knowledge)
about the most appropriate values for the parameters
from GA.
These two types of uncertainty significantly im-
pact the results and performance of a genetic algo-
rithm, as some of the uncertainty factors can affect the
quality of the result, leading to inaccurate results or a
longer computation time than necessary (Majumder
et al., 2018). In addition, uncertainty can make it dif-
ficult to compare results between different cycles of
executions of the genetic algorithm.
Therefore, it is essential to consider the types of
uncertainty and their impacts when developing and
evaluating genetic algorithms and finding strategies to
mitigate such expected impacts.
3 RELATED WORK
In this section, we presented current studies consid-
ering these types of uncertainty and their impacts on
GAs overall performance.
Some studies evaluate the behavior of genetic al-
gorithms by varying the crossover and mutation rate
as pointed out by (Hassanat et al., 2019), (Sun and Lu,
2019). In addition to these studies, some use Fuzzy
Logic such as (Fadel et al., 2021), which shows its
use in reducing GA convergence time. However, in
most of these works, there is no comparison between
the developed algorithm, the impact of selection and
substitution methods, and the characterization of un-
certainty in the genetic algorithm.
The work of (Hassanat et al., 2019) shows the im-
portance of varying the mutation rate and crossover
population size to solve various problems, demon-
strating each method’s impact in finding the optimal
solution. However, selection and substitution meth-
ods also show how they can help or hinder the search
Boosting GA Performance: A Fuzzy Approach to Uncertainty Issues Involving Parameters in Genetic Algorithms
751
for the optimal individual.
The work of (Sun and Lu, 2019) depicts that the
premature convergence of GA unexpectedly affects
the algorithm’s performance. The main reason is that
the evolution of exceptional individuals, which mul-
tiply rapidly, will lead to premature loss of popula-
tion diversity. To solve the problem, we constructed a
method to qualify the diversity of the population and
the similarity between adjacent generations.
The experiment results show that it can search for
the optimal solution for almost all reference functions
and effectively maintain the diversity of the popula-
tion. Where (Fadel et al., 2021) uses the integrated
Fuzzy Logic and GA method and uses the growth rate
function to calculate the change of maximum fitness
and average fitness between two successive genera-
tions to adapt the crossover and mutation parameters
of GA dynamically, the results showed that the pro-
posed technique gives a high and more accurate per-
formance in terms of maximum fitness and average
fitness. Despite this, no investigation shows the im-
pact of choosing the combination of selection and re-
placement methods on the algorithm’s performance
and also observes how the algorithm would behave if
we chose diversity population as a parameter in fuzzy
Logic.
In (Bajaj and Sangwan, 2019), work investigates
the effectiveness of test case prioritization based on
genetic algorithms. The work aims to find the best pa-
rameter settings and operator roles to increase the ef-
fectiveness of test case prioritization, which can save
time and resources. A strength of this work is the in-
vestigation of how parameter configuration and oper-
ator roles affect the effectiveness of test case prioriti-
zation, providing a basis for improving the effective-
ness of test case prioritization. However, a weakness
is that the work is specific to genetic algorithm-based
test case prioritization, which means that its findings
do not necessarily apply to other optimization prob-
lems. Nevertheless, fuzzy Logic considerably im-
proves the convergence time of the genetic algorithm
compared to simple GA.
In addition, it shows how easy it is to implement
and how it helps programmers who will use this tech-
nique to solve problems. However, one question re-
mained: How can the choice of types of combination
of selection and substitution methods impact the con-
vergence time and uncertainty reduction using the al-
gorithm optimized by Fuzzy Logic? We pointed out
the ease of implementing these procedures of self-
adaptation of mutation rate within the pseudo-code of
a Genetic Algorithm (see algorithm 1) and the help
that adopting these methods delivers to novice pro-
grammers in evolutionary computing.
4 METHODOLOGY
This section was divided into steps to show the treat-
ment of uncertainties presented in GA. The first
shows the environment in which we ran the algo-
rithm, the second describes a chosen use case, the
third presents the selected evaluation metrics to com-
pare the algorithms to design and measure how good
they were, and finally, it shows how a simple GA, in
operation and a GA that used the treatment of uncer-
tainty with fuzzy vagueness that was implemented in
the GA, as well as the combination of selection and
substitution parameters chosen in this work.
4.1 Test Environment
The test environment used to make the comparisons
was Google Colab (Google Computer Engine) which
has the following settings:
RAM: 12GB
HD: 108GB
We used Python version 3.7.13, and some libraries
were also used, such as:
ipython-autotime version 0.3.1 to determine the
execution time
matplotlib version 3.2.2 to display the graphs
scikit fuzzy version 0.4.2 to build the fuzzy sys-
tem
4.2 Case Study
The case study chosen in this work was OneMax,
which consists of counting the one (1) bit each chro-
mosome has and also represents the individual’s fit-
ness. Thus, the optimal binary is the individual who
has all bits in one (1). Although the solution space
or domain of the OneMax problem depends on the
length of the chromosome, an essential feature of the
OneMax domain is that all bits are unrelated (Giguere
and Goldberg, 1998). The simplicity of the OneMax
problem makes it an excellent candidate for studying
uncertainty reduction by evaluating the performance
of the simple genetic algorithm and the self-adaptive
algorithm on mutation rate and verifying selection
and replacement methods.
4.3 Evaluation Metrics
The metrics chosen in this work to perform the com-
parison of the Simple GA and the GA with the ap-
plication of Fuzzy Logic were the number of genera-
tions to find the solution to the problem, the diversity
ICAART 2024 - 16th International Conference on Agents and Artificial Intelligence
752
of the population, the execution time in seconds re-
quired for the genetic algorithm to find the optimal so-
lution. Thus, shows, in a broad way, the performance
obtained in the choice of each mutation method, se-
lection, and replacement method.
In addition, We adopted the Kruskal-Wallis hy-
pothesis test, the non-parametric test used when com-
paring three or more independent samples. It tells
us if there is a difference between the GAs that
used the combinations with substitution and selection.
For comparison between the GAs, the Dwass-Steel-
Crichtlow-Fligner (DSCF test) was used.
4.4 Algorithm Specification
This section is divided into each procedure presented
in implementing the Genetic Algorithms adopted
that are AGEE, AGETV, AGRE, AGRTV, AGFEE,
AGFETV, AGFRE, AGFRTV. To show in detail, the
choices adopted were divided into three topics, the
first showed the basis of the algorithms, that is, what
they have in common, in second showed the types of
simple GA and the nomenclature used and in the third
the types of proposed GA, the nomenclature used and
the application of fuzzy Logic.
4.4.1 Basis of Both Algorithms
It shows how the basis of the two algorithms is de-
fined, being:
The representation defined was binary.
Adopted chromosome size was twelve (12).
Population size was one hundred (100) individu-
als.
The crossover, the rate was 90%, so as not to in-
terfere with the final result.
Crossover was chosen with a cutoff point.
The algorithm has the following format, as seen in
(Jong, 2009):
4.4.2 Simple Genetic Algorithm, Without
Uncertainty Treatment
In the Simple Genetic Algorithm, variations are found
with the selection and substitution method, but the
mutation rate is fixed.
Adopted mutation rate 50
The nomenclature shown in Table 1 combines the
selection and substitution methods.
Input: Typical Parameters
Output: Final Solution Population
1: INITIALIZES population with random
candidate solutions
2: EVALUATE each candidate
3: while condition do
4: SELECT parents
5: RECOMBINE pairs of parents
6: MUTATION the resulting descendants
7: EVALUATES new candidates
8: REPLACES the individuals for the new
generation
9: end while
Algorithm 1: Pseudocode of a Typical GA.
Table 1: Nomenclature of Simple GA Types.
Substitution
Selection Elitist Life Time
Elitist AGEE AGETV
Roulette Method AGRE AGRTV
4.4.3 Genetic Algorithm with Fuzzy Logic
Approach
The GA combined with fuzzy Logic implies modifi-
cations in the methods of selection and substitution,
in addition to the definition of the implementation of
Fuzzy Logic.
Table 2 shows the terminology used in the algo-
rithms implemented with combinations of selection
and substitution methods:
Table 2: Nomeclature of the Proposed GA Types.
Substitution
Selection Elitist Life Time
Elitist AGFEE AGFETV
Roulette Method AGFRE AGFETV
4.4.4 Application of Fuzzy Logic
The population diversity is assessed by calculating the
percentage of unique individuals within the current
population, i.e., those with non-repeating genetic ma-
terial, as depicted in Equation 1. This computation re-
flects the diversity within the current population. This
calculation is performed globally at each algorithm
generation to determine the population diversity rate.
Additionally, the individual’s score within the popu-
lation is computed to be used as input in Fuzzy rules,
thereby determining the mutation percentage (MP).
PD =
UniqueIndividuals
Population
x100 (1)
Boosting GA Performance: A Fuzzy Approach to Uncertainty Issues Involving Parameters in Genetic Algorithms
753
Twenty-five rules for the OneMax prob-
lem were considered in the application using
the triangular membership function representa-
tion. The rules have the following antecedent
values of individual’s adaptation and popula-
tion diversity (PD), both with the fuzzy set
0
verybad
0
,
0
bad
0
,
0
medium
0
,
0
good
0
, and
0
verygood
0
as represented in Figure 1 and as a consequence the
individual’s MP, which is represented by the fuzzy set
0
verylow
0
,
0
low
0
,
0
medium
0
,
0
high
0
,
0
veryhigh
0
which is
depicted in Figure 2 [ 0,100].
Figure 1: Population Quality and Diversity.
Figure 2: Mutation Percentage Graph.
The rules chosen in the fuzzy system were ex-
pressed as follows:
5 RESULTS AND DISCUSSION
To compare uncertainty issues and their treatment
performance, we executed each algorithm a hundred
(100) times; that is, each execution means that the al-
gorithm was started and just stopped when it found
the solution to the problem. When found, the data is
stored, and the system is restarted. This looping hap-
pens until the 100 runs are completed, and then, the
extraction of the results is performed. Another critical
factor was the division into two sections, one with the
Table 3: Fuzzy Rules.
results of the Simple GA and another with the results
of the Proposed GA.
5.1 Results Obtained in Simple GA,
Without Uncertainty Treatment
First, simple genetic algorithms’ convergence time
and their variations were analyzed.
Figure 3 shows the time for each of the 100 runs
that each algorithm obtained, that is, the time each run
took to find the best solution to the OneMax problem.
Given the above, it is remarkable the difference that
exists in the execution time of the algorithms.
The AGRE algorithm, on average, obtained the
worst performance, about 22.80 seconds (s), com-
pared to the other simple genetic algorithms, which
obtained AGEE 21.60 s, AGETV 0.30 s, and AGRTV
0.28 s.
One of the reasons analyzed for this difference in
time of the simple GA that chose the Elitist substitu-
tion about the simple GA with the lifetime substitu-
tion was the delay of convergence of the population
as seen in (Kim and Han, 2000) due to the size of the
population that directly impacts on its convergence
time, that is, the substitution by Life Time maintains
ICAART 2024 - 16th International Conference on Agents and Artificial Intelligence
754
Figure 3: Execution Time Simple GA.
a more significant amount of living individuals per
time interval, unlike the Elitist that in each iteration
eliminates the individuals with a lower score, impair-
ing the emergence of the optimal solution. Another
critical point is that the roulette wheel method allows
the crossover to individuals with different configura-
tions of their chromosomes, bringing them good or
bad, which provides a larger sweep in the Exploita-
tion search. At the same time, the Electoral selection
can be stuck at the local optimum.
Figure 4: Population Diversity by Generation Simple GA.
In addition, the diversity of the population present
in a run was verified, as shown in Figure 4. It was
observed that the Simple Genetic Algorithms with
Elitism in the present replacement diversity, in the be-
ginning, were equal to the simple GA with replace-
ment by lifetime, but over the generations, this di-
versity changes because the algorithms AGETV and
AGRTV have a larger population, but the lifetime set
in each generation is different, thus implying a more
significant amount of living individuals, but when the
maximum lifetime is reached this generation is de-
stroyed, which helps in the high diversity. On the
other hand, the AGEE and AGRE algorithms have a
reduced population, but the amount of different indi-
viduals within it is not large by keeping optimal indi-
viduals, which sometimes have the same genetic ma-
terial.
Figure 5: Result of the Generation by Run.
Finally, Figure 5 shows the number of genera-
tions needed to find the optimal individual in each
run, which averaged AGRTV 41.48, AGRE 5253.03,
AGEE 4798.50 and AGETV 41.42. Being high-
lighted, the Simple GA is implemented with the life-
time, which can find the optimal individual first and
with a small number of iterations.
5.2 Results Obtained in the Proposed
GA with Fuzzy Approach
The tests performed on the Simple Genetic Algorithm
were also performed on the Genetic algorithm com-
bined with Fuzzy Logic with its variations on the
substitution and selection method to be able to make
comparisons between them.
Figure 6: Proposed GA Runtime.
The AGFRE algorithm had the worst perfor-
mance, averaging about 6.52 s compared to the
other genetic algorithms combined with Fuzzy Logic,
which had AGFETV 0.16s, AGFRTV 0.30s, and
AGFEE 0.23s.
The self-adaptive genetic algorithms that used the
method of Lifetime Replacement have an increased
population, which impacts reducing the convergence
time of the GA. In addition, the self-adaptive GA
maintains a high search in the Exploration approach,
Boosting GA Performance: A Fuzzy Approach to Uncertainty Issues Involving Parameters in Genetic Algorithms
755
i.e., effectively sweeps a more significant number of
search spaces; this occurs because it can infer an ap-
propriate mutation rate for each chromosome in the
population, taking into account the diversity of the
population and the quality of the chromosome, which
interferes with faster convergence.
Figure 7: Population Diversity by Generation Proposed GA.
There was a check on the diversity of the popu-
lation present in a run, as shown in Figure 7. It was
observed that the Self-Adaptive Genetic Algorithms
with Electism in the substitution present diversity, in
the beginning, equal to that of the Self-Adaptive GA
with replacement by Life Time, but during the gener-
ations, this diversity diverges because the algorithms
AGFETV and AGFRTV have a larger population and
their diversity as well. On the other hand, the algo-
rithms AGFEE and AGFRE have a reduced popula-
tion, which impacts an increasing diversity through-
out the generations; that is, there is a greater chance
of having individuals with the same genetic material.
Figure 8: Generation of Each Proposed GA Run.
Figure 8 illustrates the number of generations
needed to find the best-fit chromosome, which had an
average of AGFRTV 7.92, AGFRE 356.84, AGFEE
3.76, AGFETV 3.40 generations. The algorithms that
stood out were AGFEE and AGFETV, which had the
Elitist selection method. It shows the importance of
correctly defining the mutation rate and the choice of
selection and substitution methods.
5.3 Hypothesis Testing
The Kruskal-Wallis test is a non-parametric test used
when comparing three or more independent samples.
It indicates if there is a difference between at least two
of them (Rumsey, 2015).
Therefore, the number of generations needed to
find the optimal individual per run was selected for
analysis. First, the following hypotheses were raised
to check if there is a difference in the choice between
the selection and replacement methods:
H
0
: There is no statistically significant difference
between the group means.
H
1
: There is a difference between group means.
A significance level of 5% was chosen. The test
result was less than 0.001 s; that is, the result is lower
than the significance level, and there is a statistical
difference. Table 4 shows the comparison between
each of the Simple GA types.
Table 4: Kruskal-Wallis Test Simple GA.
W p
AGEE AGETV -17.261 0.001
AGEE AGRE 0.734 0.955
AGEE AGRTV -17.257 0.001
AGETV AGRE 17.225 0.001
AGETV AGRTV 1.294 0.797
AGRE AGRTV -17.240 0.001
Table 4 shows no difference between the AGEE
and AGRE methods, and the same is true for AGETV
and AGRTV. However, the Life Time Replacement
method causes a statistically relevant difference in the
cases.
The same test was performed with the GAs that
used Fuzzy Logic, adopting a significance level of
5%. Again, the result was less than 0.001 s; that is,
there is a statistical difference between the groups.
Table 5 shows the comparison between the groups.
Table 5: Kruskal-Wallis Test GA with Fuzzy Logic.
W p
AGFEE AGFETV -0.142 0.987
AGFEE AGFRE 17.317 0.001
AGFEE AGFRTV 6.721 0.001
AGFETV AGFRE 17.308 0.001
AGFETV AGFRTV 6.9787 0.001
AGFRE AGFRTV -17.283 0.001
In Table 5, there is no difference between the
AGFEE and the AGFETV methods, whereas there is
ICAART 2024 - 16th International Conference on Agents and Artificial Intelligence
756
a statistically significant difference between the other
methods.
6 CONCLUSION AND FUTURE
WORK
In this article, we present an approach that addresses
types of uncertainty inherent in the development and
execution of genetic algorithms. Experimentally, we
applied the proposed method with the fuzzy approach
and different selection and substitution methods. We
showed that the genetic algorithm combined with
the conceptual vagueness treatment, using the elitist
selection and the lifetime substitution method, pre-
sented the best results compared to the other combi-
nations. In this sense, this work highlights the rel-
evance of this approach to improve the performance
of genetic algorithms and also shows the reduction of
uncertainty found in genetic algorithms through im-
plementing a self-adaptive algorithm that helps and
facilitates the search for the global optimum. This
contribution is a significant step towards improving
performance and knowledge representation in GA.
In our approach, Fuzzy Logic was used in the pro-
posed algorithm in a simple way to implement, which
confirms the adoption of these modifications in its
implementation, showing the importance of modifi-
cation in the current population-sensitive interactive
mutation rate.
In addition, it was also shown the importance of
testing with different methods of substitution and se-
lection because the reduction in convergence time is
significant in the OneMax problem, this difference in
most cases tested being greater than 90%, i.e., the
methods that used Time of Life substitution had an
advantage. On the other hand, the methods that used
Elitism had an increasing diversity along the genera-
tions and a long time of convergence.
Using Fuzzy has a downside linked to the algo-
rithm’s convergence time; a larger population results
in longer computational times. This study is lim-
ited by analyzing only a single test case, the Onemax
problem, diminishing the generalization of diversity
understanding for uncertainty reduction in the GA.
In future work, we intend to apply the proposed
GA to other classes of problems and analyze how this
algorithm behaves. In this way, it facilitates the devel-
oper in deciding which problems the adoption of the
self-adaptive genetic algorithm is recommended. For
example, in the Traveling Salesman Problem (TSP)
function Optimization, the methodology adopted in
this work focuses on the variation of the mutation
rate and variation of the substitution and selection
method to corroborate the results obtained in this
work. Linked to this, analyze in more detail in Fuzzy
Logic the consequences of changing the mutation rate
in different functions of representation of knowledge
as the Trapezoidal, Gaussian.
Another thread for future work is to verify if the
methodology applied here can be extended to all rates,
such as crossover and population size variation. Thus,
the importance of changing the parameters during the
execution of the genetic algorithm iterative can be
seen.
REFERENCES
Bajaj, A. and Sangwan, O. P. (2019). Study the impact of
parameter settings and operators role for genetic algo-
rithm based test case prioritization. In Proceedings of
International Conference on Sustainable Computing
in Science, Technology and Management (SUSCOM),
Amity University Rajasthan, Jaipur-India.
ˇ
Crepin
ˇ
sek, M., Liu, S.-H., and Mernik, M. (2013). Explo-
ration and exploitation in evolutionary algorithms: A
survey. ACM computing surveys (CSUR), 45(3):1–33.
Fadel, I. A., Alsanabani, H.,
¨
Oz, C., Kamal, T., Iskefiyeli,
M., and Abdien, F. (2021). Hybrid fuzzy-genetic al-
gorithm to automated discovery of prediction rules.
Journal of Intelligent & Fuzzy Systems, 40(1):43–52.
Giguere, P. and Goldberg, D. E. (1998). Population siz-
ing for optimum sampling with genetic algorithms: A
case study of the onemax problem. Genetic Program-
ming, 98:496–503.
Hassanat, A., Almohammadi, K., Alkafaween, E.,
Abunawas, E., Hammouri, A., and Prasath, V. S.
(2019). Choosing mutation and crossover ratios for
genetic algorithms—a review with a new dynamic ap-
proach. Information, 10(12):390.
Jong, K. D. (2009). Evolutionary computation. Wiley
Interdisciplinary Reviews: Computational Statistics,
1(1):52–56.
Kim, K.-j. and Han, I. (2000). Genetic algorithms approach
to feature discretization in artificial neural networks
for the prediction of stock price index. Expert systems
with Applications, 19(2):125–132.
Majumder, S., Saha, B., Anand, P., Kar, S., and Pal, T.
(2018). Uncertainty based genetic algorithm with
varying population for random fuzzy maximum flow
problem. Expert Systems, 35(4):e12264.
Rumsey, D. J. (2015). U Can: statistics for dummies. John
Wiley & Sons.
Sun, N. and Lu, Y. (2019). A self-adaptive genetic algo-
rithm with improved mutation mode based on mea-
surement of population diversity. Neural Computing
and Applications, 31(5):1435–1443.
Boosting GA Performance: A Fuzzy Approach to Uncertainty Issues Involving Parameters in Genetic Algorithms
757