Selection of Representative Instances using Ant Colony: A Case Study in
a Database of Children and Adolescents with
Attention-Deficit/Hyperactivity Disorder
Henrique R. Hott
1
, Caroline R. Jandre
1
, Pedro H. S. Xavier
1
, Amal Miloud-Aouidate
2
,
D
´
ebora M. Miranda
3
, Mark A. Song
1
, Luis E. Z
´
arate
1
and Cristiane N. Nobre
1
1
Department of Computer Science, Pontifical Catholic University of Minas Gerais University, Brazil
2
University of Sciences and Technology Houari Boumediene, Algeria
3
Department of Pediatrics, Federal University of Minas Gerais, Minas Gerais, Brazil
Keywords:
ADHD, Attention-Deficit/Hyperactivity Disorder, Instance Selection, Ant Colony.
Abstract:
Instance Selection (IS) helps select the most notable instances from the database, improving its characteriza-
tion and relevance. In this context, this article applies the IS, using the Ant Colony Optimization (ACO) heuris-
tic, to obtain more efficient classification models in the identification of school performance, in arithmetic,
writing, and reading, of children and adolescents with Attention-Deficit/Hyperactivity Disorder (ADHD),
characterized by excessive symptoms of inattention, hyperactivity, and impulsivity. The Random Forest, Neu-
ral Networks, KNN, and CART classifiers were used to evaluate the performance of the selection performed
by the ACO method. With the ACO, it was possible to obtain a gain of 20 percentage points with the KNN
(K = 1), in arithmetic, in the metric F-measure, referring to the upper class, the minority class. The results
achieved show the excellent efficiency of the ACO in this study.
1 INTRODUCTION
Data Mining (DM) is the process of obtaining prac-
tical knowledge from a sufficiently large dataset in
an automated or semi-automated manner. There are
many processes in this field, mainly supervised learn-
ing, whose results are affected by the quality or size
of the dataset used for knowledge extraction. Further-
more, many algorithms can become unfeasible in the
face of a large data set.
The Data Reduction (DR) area defines a series of
pre-processing tasks to handle the abovementioned
problem. The primary function of DR techniques is
to reduce the size of a dataset by selecting the most
representative information (Pyle, 1999). Therefore,
the main objective of DR techniques is to choose a
representative subset of data that offers results with
performance equal or better to the same experiments
without the data reduction step (Garcia et al., 2014).
Instance Selection (IS) represents one of the main
tasks in DR (Liu and Motoda, 2001). The role of IS
is to select the most significant rows in the dataset.
One usage of IS algorithms is to improve the charac-
terization of instances. Consequently, it can be help-
ful for health-oriented databases applications such
as Attention-Deficit/Hyperactivity Disorder (ADHD),
creating a more representative model.
Attention-Deficit/Hyperactivity Disorder
(ADHD) is a neuropsychiatric disorder charac-
terized by the symptomatic triad: inattention,
hyperactivity, and impulsivity, which is excessively
manifested through behavior, and speech. In addition
to these central behaviors, people with ADHD usu-
ally have difficulty organizing their daily tasks and
regulating their emotions. The worldwide prevalence
of ADHD is around 5.3% in children and adolescents
and 2.5% in adults (da Silva et al., 2020; Retz et al.,
2020). ADHD can harm the diagnosed individual’s
social, educational, and family life (Mattos, 2015).
Decreased academic performance and success, social
rejection, and relationship difficulties are often
related to the disorder, which leads to considerable
educational and social losses (Rangel J
´
unior and
Loos, 2011).
Thus, the objective of this work is to apply the IS
using the Ant Colony Optimization heuristic to ob-
Hott, H., Jandre, C., Xavier, P., Miloud-Aouidate, A., Miranda, D., Song, M., Zárate, L. and Nobre, C.
Selection of Representative Instances using Ant Colony: A Case Study in a Database of Children and Adolescents with Attention-Deficit/Hyperactivity Disorder.
DOI: 10.5220/0010843000003123
In Proceedings of the 15th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2022) - Volume 5: HEALTHINF, pages 103-110
ISBN: 978-989-758-552-4; ISSN: 2184-4305
Copyright
c
 2022 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
103

tain more efficient classification models in identify-
ing the school performance of people with ADHD -
Attention-Deficit/Hyperactivity Disorder. In this con-
text, we analyzed a database containing the perfor-
mance of 266 children and adolescents in the follow-
ing subjects: Arithmetic, Writing, and Reading. Fur-
thermore, the Random Forest, Neural Networks, K-
Nearest Neighbors (KNN) and Classification and Re-
gression Trees (CART) classifiers were used to eval-
uate the efficiency of the selection performed by the
Ant Colony method.
The article follows the following structure: in Sec-
tion 2, the theoretical foundation is presented, which
brings the main concepts related to work. The works
related to the theme are covered in Section 3. Section
4 presents the methodology used, with a detailed de-
scription of the database and the pre-processing steps.
Section 5 gives the discussions regarding the results
found. Finally, in Section 6, the final considerations
in this article are exposed.
2 BACKGROUND
2.1 Attention-Deficit/Hyperactivity
Disorder
The characteristics of ADHD are related to a dysfunc-
tion of the brain’s frontal lobe neurons, and its com-
plex etiology stems from genetic and environmental
factors. Its symptoms vary according to the stage of
development of the disorder (da Silva et al., 2020;
Jennum et al., 2020). Its diagnosis is made through
symptomatology since there are no specific biomark-
ers that indicate it. Therefore, it is essential to dif-
ferentiate the symptoms of the disorder with behavior
characteristic of children of active age, such as exces-
sive noise and running around, or even with symp-
toms of other disorders, such as anxiety and mood
(da Silva et al., 2020).
ADHD individuals may have the predominantly
inattentive subtype, which concerns only characteris-
tics related to inattention. Or the predominantly hy-
peractive/impulsive subtype, which is characterized
only by hyperactive/impulsive behaviors. Or the com-
bination of both, which is the most prevalent of the
subtypes. In this case, the individual has both inat-
tentive and hyperactive/impulsive characteristics. It
happens in about 62% of people with ADHD (APA
et al., 2014; Cardoso et al., 2018). ADHD is usu-
ally noticed when the individual is of school age since
it is the phase in which concentration is most neces-
sary. However, its symptomatic signs may extend into
adulthood (Santos and Vasconcelos, 2010).
The therapeutic follow-up of people with ADHD
includes psychosocial and drug interventions. The
symptoms of inattention, hyperactivity, and impul-
sivity can significantly impact academic development
and the areas of neurodevelopment, psychosocial in-
teraction, adaptive functioning, and emotional self-
regulation of the individual. All of this can contribute
to low self-esteem and the emergence of other diffi-
culties in your life (Maia and Batista, 2021). Thus, the
earlier diagnosis of ADHD and the implementation of
therapeutic measures, the smaller the negative impact
that the disorder may have on people with the disor-
der and its surrounding (Moreira and Barreto, 2017;
Muzetti and de Luca Vinhas, 2017).
2.2 Instance Selection with Ant Colony
Instance Selection is one of the most critical pre-
processing tasks that play a central role in the DR
area. The main objective of IS is to find the mini-
mum cardinality subset of the original samples whose
execution of a Data Mining (DM) algorithm has a
performance equal to or with less deterioration than
compared with using the complete dataset (Liu and
Motoda, 2001).
The IS problem is classified as an NP-hard prob-
lem, as it requires a thorough search of every pos-
sibility to find the best solution. Dealing with this
high complexity requires a lot of research effort to
find heuristics or approximate solutions that obtain
acceptable results considering a reasonable time and
computational resources.
The traditional IS approach in the literature uses
neighborhood criteria to eliminate noisy or redundant
instances. The traditional technics are focused on im-
proving the performance of classifiers based on the
nearest neighbors rule (Garcia et al., 2012). In this
approach, an example is retained if it disagrees with
the established metrics of similarity.
In addition to the classical approach, the use of
constructive heuristics has also been successfully ap-
plied in IS, with emphasis on the evolutionary algo-
rithms (EAs) (Derrac et al., 2010). However, in this
work, we will use the well-known Ant Colony Opti-
mization (ACO) (Dorigo and St
¨
utzle, 2004), another
type of constructive heuristic not much explored for
the IS task.
ACO is a search meta-heuristic proposed in
Dorigo and Di Caro (1999), designed to tackle com-
binatorial problems, inspired by the foraging behav-
ior observed in ant colonies in finding the shortest
path between a food source and a nest. The main
idea is that the self-organization principles that allow
HEALTHINF 2022 - 15th International Conference on Health Informatics
104

the highly coordinated behavior of natural ants can
be exploited to coordinate groups of artificial agents
that collaborate to solve computational problems. A
pseudo-code of a basic ACO algorithm is described in
Algorithm 1.
Algorithm 1: Basic ACO Pseudocode.
input: Any Combinatorial Problem
InitializePheromoneValues(T );
s
bs
← NULL;
while terminination conditions not met do
S
iter
←
/
0;
for j = 1, ..., n
a
do
s ← ConstructSolution(T );
if s is a valid solution then
if f (s) > f (s
bs
) or (s
bs
= NU LL)
then
s
bs
← s;
end
S
iter
← S
iter
∪ {s};
end
end
ApplyPheromoneU pdate(T, S
iter
, s
bs
);
end
output: The best-so-far solution s
bs
ACO agents (artificial ants) represent stochastic
construction procedures that incrementally generate
the solution by adding opportunely defined solution
components. In essence, an ACO algorithm initially
transforms the search space into a graph linking the
possible solution components. The colony then nav-
igates over this graph building its candidate solution.
The probability of an ant selecting the path to its
next solution component is based on two properties:
the heuristic advantage and the amount of pheromone
present in it.
The core of an ACO algorithm is the T pheromone
model used. The model defines τ
i j
parameters for
every possible path of the search space. These pa-
rameters represent the knowledge accumulated by the
colony. A higher value in τ
i j
indicates to an ant that
choosing that path {i, j} will lead to a high-quality
region.
An ant increments the value of the pheromone pa-
rameters belonging to its current route. Consequently,
a path that more agents choose has its pheromone val-
ues reinforced, increasing the probability that more
ants will choose it in future iterations. This behavior
implicitly deduces that good solutions are made with
good solution components.
At the end of each iteration of an ACO algorithm
occurs a simulation of pheromones evaporation by the
environment. In this step, the amount of pheromones
deposited in previous iterations reduces at a defined
rate. Such a process helps to reduce the chance that
the algorithm will early converge on local maxima.
As mentioned at the beginning of the section, IS
is an NP-Hard problem that requires an exhaustive
search to find the best set. In this way, we can also
classify an IS as a combinatorial problem.
At the end of each iteration of an ACO algorithm
occurs a simulation of pheromones evaporation by the
environment. In this step, the amount of pheromones
deposited in previous iterations reduces at a defined
rate. Such a process helps to reduce the chance that
the algorithm will early converge on local maxima.
As mentioned at the beginning of the section, IS
is an NP-Hard problem that requires an exhaustive
search to find the best set. In this way, we can also
classify an IS as a combinatorial problem.
Because of this property, we can easily define an
ACO algorithm for the IS task. The input set instances
can create the search graph that combines the decision
components that represent adding a sample to the re-
duced set. In this way, the colony navigates the graph
creating subsets of examples. A DM algorithm or an-
other specified metric evaluates the subsets by the per-
formance obtained.
3 RELATED WORKS
The IS approach with an ACO metaheuristic remains
less explored in the literature than other construc-
tive heuristics. Among the studies, two stand out for
providing us with new ACO models for the IS task
(ACO-IS).
Anwar et al. (2015) proposed a new ACO algo-
rithm for the IS task called ADR-Miner. The new
approach aims to improve the accuracy of the clas-
sification model provided as a parameter by selecting
the most representative samples. The ADR-Miner has
been validated in 20 public databases offering promis-
ing results that surpass classical IS approaches. Fur-
thermore, in (Salama et al., 2016) the ADR-Miner
was enhanced to perform the attribute selection task
together.
The Ant-IS algorithm used in this work is an
ACO-IS algorithm of the condenser type proposed
in Miloud-Aouidate and Baba-Ali (2015) that uses
ant colony principles to perform the IS. In addition,
the technique uses a nearest neighbor classifier to
assess the similarity between subsets generated by
the colony. In experiments carried out with public
databases, Ant-IS surpassed results obtained for clas-
sification models without IS.
Selection of Representative Instances using Ant Colony: A Case Study in a Database of Children and Adolescents with
Attention-Deficit/Hyperactivity Disorder
105

On the educational theme, Li (2019) proposes
an adaptive online learning model based on the ant
colony algorithm, which seeks to better respond to
people’s demands for multimedia learning. Appropri-
ate learning paths must match users’ personalized in-
formation, including personal education background
and learning style, catering to different student prefer-
ences, tastes, and knowledge levels, without the need
for them to be aware of it. According to the author, the
adaptive learning system based on an ant colony algo-
rithm can help teachers develop personalized courses
for different students and provide suitable learning
objects, helping to improve students’ academic per-
formance and learning efficiency.
In the health area, the ant colony can help in the
diagnosis of diseases. Selcuk and Alkan (2019) used
the ant colony algorithm to aid in the accurate, ef-
fective, and automatic detection of microaneurysms,
which are difficult to detect in color fundus images in
the early stages of diabetic retinopathy. Accurate de-
tection of these lesions is critical in the early diagno-
sis of this disease. The same procedure was also ap-
plied to five different image processing and clustering
algorithms for performance comparison. The results
show that the ant colony-based method proposed in
this work successfully detects microaneurysms even
in low-quality images, facilitating early diagnosis by
specialists.
The related works help prove the effectiveness
of using the ant colony, validating its application in
work. However, the differential of this article is the
use of the ant colony on a database focused on the
school performance of students with ADHD.
4 MATERIALS AND METHODS
4.1 Description of the Database
The database was made available by a Brazilian Uni-
versity based on individual and family responses
present in questionnaires. The sample is composed
of students aged between 6 and 18 years old, with and
without a diagnosis of ADHD. There are data related
to health, financial conditions, parental care and ed-
ucation amongst others, in addition to the grades for
each student’s arithmetic, writing, and reading tests.
As the objective is to identify, from a posterior
perspective, the academic profile of students in each
discipline. The original database was divided into
three, in which each class is represented by a disci-
pline. After the split, there were some pre-processing
steps:
• Exclusion of instances that had no value in the re-
spective class. Table 1 presents the total number
of instances in each discipline and also the num-
ber of instances distributed in High and Low.
Table 1: Number of instances.
Discipline
Total instances by performance Total
instancesHigh Low
Arithmetic 70 189 259
Writing 59 203 262
Reading 47 177 224
• Filling in the missing data by the average, in nu-
merical data, or by mode, in categorical features.
• Binarization of non-ordinal nominal features, that
is, they were coded as the presence or absence of
the characteristic.
• Attribute selection through the Genetic Algorithm
(GA). In this step, the Non-dominated Sorting Ge-
netic Algorithm-II (NSGA-II) algorithm was cho-
sen to find the best subset of features maximizing
its fitness, in this case, the F-measure. To mea-
sure the F-measure, the classifier KNN was used,
varying the value of the K in the range of [1-10].
The number of GA generations was the stopping
criterion in analyzing the values that best fit the
algorithm parameters. For each set of parame-
ters, 10 different random seeds were used. The
GA was implemented in the Python language, us-
ing the DEAP library, available from Universit
´
e
Laval (Fortin et al., 2012). Figure 1 shows which
features will remain in each discipline after select-
ing the GA.
4.2 Ant Colony Description
The Ant-IS algorithm proposed in Miloud-Aouidate
and Baba-Ali (2015) was used in the present work.
This technique aims to condense as many instances
as possible with a minimal negative impact on the
performance of classification models. This method
uses the ACO principles (Dorigo and St
¨
utzle, 2004)
to perform the instance selection. In addition, the al-
gorithm internally uses a nearest neighbor classifier
(Cover and Hart, 1967) to validate the subsets selected
by the colony.
In this way, each ant in the colony performs a par-
allel search in its neighborhood for nearby instances,
which are also as close as possible to the initial search
point. The IS was performed during neighborhood
analysis, in which an ant randomly decides to include
or not an instance in its reduced set.
There is a particularity between Ant-IS and other
ACO methods. The number of ants used by the colony
HEALTHINF 2022 - 15th International Conference on Health Informatics
106

Figure 1: Features selected by the GA in each discipline.
is not defined as a user parameter, but rather one
ant per instance is created. According to the algo-
rithm’s authors, given each instance as initial points,
all path probabilities are explored without restricting
the search to an initial condition.
The Ant-IS implementation provides the follow-
ing parameters for the user: the parameter τ
init
de-
fines the initial amount of pheromone in each search
path; the Q parameter that controls the amount of
pheromone deposited by an agent; and parameter
ρ ∈ [0, 1] which represents an evaporation rate of
pheromones at the end of an iteration.
The Euclidean distance was used as a similarity
metric to calculate the distance between the set infor-
mation. We define the remaining parameters in our
experiments as follows: τ
init
= 1, Q = 1 and ρ = 0.1.
4.3 Machine Learning Methods Used
Four classification algorithms were used to evaluate
the performance of Ant-IS: 1NN; the CART Decision
Tree algorithm; the Neural Network Backpropagation
algorithm; and Random Forest (RF), a set of methods
that uses a set of decision trees to perform its classifi-
cation.
For the algorithms mentioned above, the imple-
mentations provided by Scikit-learn (Pedregosa et al.,
2011) were used. It is worth emphasizing that Scikit-
learn uses a more performing version of the CART al-
gorithm called by the DecisionTreeClassifier library.
4.4 Model Quality Assessment Metrics
Precision, Recall, and F −measure metrics were used
to determine the quality of the models. The Precision
1
metric concerns the percentage of instances classified
correctly in a class out of all those that were classified
in the class. The Recall
2
refers to the percentage of
instances of a class that were correctly predicted to
belong to the class. Already the metric F-measure
3
represents the harmonic mean between Precision and
Recall.
1
Precision =
V P
V P+FP
2
Recall =
V P
V P+FN
3
F −Measure =
2×Recall×Precision
Recall+Precision
Selection of Representative Instances using Ant Colony: A Case Study in a Database of Children and Adolescents with
Attention-Deficit/Hyperactivity Disorder
107

Given the stochastic nature of the ACO algo-
rithms, the experiments were performed differently
for the Ant-IS collected sets. In this case, the ex-
periment was done using the 10-fold stratified cross-
validation proposed in Salama et al. (2016). In this
process, the cross-validation procedure is repeated ten
times. In addition, a new subset of instances is gen-
erated on each model run, and the test result is the
average result of all runs.
5 RESULTS AND DISCUSSIONS
Figure 2 presents the graph of the results obtained
in experiments. Each column referred to the experi-
ments performed with whole datasets and applied the
instance selection with ANT-IS. Furthermore, the re-
sults for each metric are presented by discipline.
Analyzing the results, it is possible to observe an
improvement in most metrics evaluated in the Ant-IS
sets. In general, this enhances the ability of Ant-IS to
select the most representative from the database. In
few cases, like the 1NN algorithm for the high class,
a reduction in the evaluation metrics was observed
with the selected subsets. In this case, there was a
reduction of 5 percentage points with the application
of Ant-IS. An important observation is about the neu-
ral network, which obtained much lower results than
other classifiers. We believe that a better adjustment
of hyperparameters can improve the performance of
the algorithms.
The most significant contribution of this work can
be seen in the improvement under the minority class
(high). Overall, the performance of the ’high’ class
with Ant-IS increased significantly, with a gain of up
to 24.06 percentage points considering the F-measure
metric, using the 1NN in the writing database. Thus,
the experiment results indicate that Ant-IS made it
possible to represent the database better.
The CART and RF models that use decision trees
to carry out the classification were obtained on the’
high’ class learning. CART gained around 23 per-
centage points of accuracy in the arithmetic discipline
and 10 in the writing discipline. Similarly, the RF
algorithm increased all its high class metrics for all
subjects, with a gain ranging from 11.1 to 18.44 per-
centage points.
Regarding the low class, only in the 1NN algo-
rithm was there a loss of 9 percentage points in the
F-measure metric in the arithmetic prediction topic.
However, in writing and reading subjects, using the
1NN, and in all other algorithms, in all disciplines, the
result was either very close to that achieved without
the use of Ant-IS or better. In the CART algorithm,
for example, in arithmetic prediction, there was a gain
of 7.41 percentage points in the F-measure metric of
this class.
Besides, we can note that the application of Ant-
IS in the database contributed to a more effective pre-
diction of student performance, especially regarding
high performance, which is represented by the minor-
ity class. This makes the knowledge acquired through
the most selected instance groups valid.
6 FINAL CONSIDERATIONS
The objective of this work, as presented, was to carry
out the selection of instances using Ant Colony to ob-
tain more efficient classification models in identifying
the school performance of children and adolescents
with ADHD.
The early identification of possible school diffi-
culties of students with ADHD can positively help
the prognosis of educational and social interventions.
In addition, finding both the characteristics that lead
to low and high performance supports the creation of
more effective mediations that minimize the effects of
elements (be they family, social, among others) that
can lead to low performance and enhance those that
help in achieving high performance.
Thus, the results presented effectively contribute
to achieving this early identification. Furthermore,
Ant-IS was essential in improving high and low-
performance prediction, making it possible to later
extract reliable rules from this database. Through
these rules, the standard that governs the low or high
performance in arithmetic, writing, and reading, of
students with ADHD can be evaluated, and, with this,
more targeted actions can be taken.
As a proposal for future work, we intend to an-
alyze the most selected instances by class and val-
idate these characteristics with an expert. Further-
more, evaluating other pheromone models for Ant-IS
can improve the results obtained.
ACKNOWLEDGEMENTS
The authors thank the National Council for Scientific
and Technological Development of Brazil (CNPq),
the Coordination for the Improvement of Higher Ed-
ucation Personnel - Brazil (CAPES), the Founda-
tion for Research Support of Minas Gerais State
(FAPEMIG) and Pontifical Catholic University of Mi-
nas Gerais (PUC Minas).
HEALTHINF 2022 - 15th International Conference on Health Informatics
108

Figure 2: Performance obtained in each experiment per metric.
REFERENCES
Anwar, I. M., Salama, K. M., and Abdelbar, A. M. (2015).
Instance selection with ant colony optimization. Pro-
cedia Computer Science, 53:248–256. INNS Confer-
ence on Big Data 2015 Program San Francisco, CA,
USA 8-10 August 2015.
APA et al. (2014). DSM-5: Manual diagn
´
ostico e estat
´
ıstico
de transtornos mentais. Artmed Editora.
Cardoso, L., Mollica, A. M. V., Sales, A. M., and Ara
´
ujo,
L. C. (2018). O l
´
udico e a aprendizagem de crianc¸as
com transtorno de d
´
eficit de atenc¸
˜
ao/hiperatividade.
Revista Cient
´
ıfica FAGOC-Multidisciplinar, 3(2):33–
41.
Cover, T. and Hart, P. (1967). Nearest neighbor pattern clas-
sification. IEEE Transactions on Information Theory,
13(1):21–27.
da Silva, M. L. V., de Sousa Soares, N., da Costa Soares,
R. S., de Andrade, M. K. B., Ramos, Z. S., dos San-
tos Abreu, I., Le
˜
ao, G. K. S. A., and de Freitas Soares,
F. A. (2020). Abordagens em sa
´
ude mental em
pessoas com transtorno de d
´
eficit de atenc¸
˜
ao com
hiperatividade (TDAH): uma revis
˜
ao integrativa. Re-
search, Society and Development, 9(8):e628985933–
e628985933.
Selection of Representative Instances using Ant Colony: A Case Study in a Database of Children and Adolescents with
Attention-Deficit/Hyperactivity Disorder
109

Derrac, J., Garc
´
ıa, S., and Herrera, F. (2010). A survey
on evolutionary instance selection and generation. In-
ternational Journal of Applied Metaheuristic Comput-
ing, 1(1):60–92.
Dorigo, M. and Di Caro, G. (1999). Ant colony optimiza-
tion: a new meta-heuristic. In Proceedings of the 1999
Congress on Evolutionary Computation-CEC99 (Cat.
No. 99TH8406), volume 2, pages 1470–1477.
Dorigo, M. and St
¨
utzle, T. (2004). Ant Colony Optimiza-
tion. Bradford Company, USA.
Fortin, F.-A., De Rainville, F.-M., Gardner, M.-A. G.,
Parizeau, M., and Gagn
´
e, C. (2012). DEAP: Evolu-
tionary algorithms made easy. Journal of Machine
Learning Research, 13(1):2171–2175.
Garcia, S., Derrac, J., Cano, J., and Herrera, F. (2012).
Prototype selection for nearest neighbor classifica-
tion: Taxonomy and empirical study. IEEE Transac-
tions on Pattern Analysis and Machine Intelligence,
34(3):417–435.
Garcia, S., Luengo, J., and Herrera, F. (2014). Data Pre-
processing in Data Mining. Springer Publishing Com-
pany, Incorporated.
Jennum, P., Hastrup, L. H., Ibsen, R., Kjellberg, J., and Si-
monsen, E. (2020). Welfare consequences for people
diagnosed with attention deficit hyperactivity disorder
(ADHD): A matched nationwide study in denmark.
European Neuropsychopharmacology, 37:29–38.
Li, R. (2019). Adaptive learning model based on ant colony
algorithm. International Journal of Emerging Tech-
nologies in Learning, 14(1):49–57.
Liu, H. and Motoda, H. (2001). Instance Selection and Con-
struction for Data Mining. Kluwer Academic Publish-
ers, USA.
Maia, G. B. C. and Batista, K. G. S. (2021). Avaliac¸
˜
ao
neuropsicol
´
ogica das func¸
˜
oes executivas e da
atenc¸
˜
ao em crianc¸as com transtorno do d
´
eficit de
atenc¸
˜
ao/hiperatividade - TDAH. Humanas em
Perspectiva, 3:41–61.
Mattos, P. (2015). No mundo da lua: perguntas e respostas
sobre transtorno do d
´
efict de atenc¸
˜
ao com hiperativi-
dade em crianc¸as, adolescentes e adultos. Associac¸
˜
ao
Brasileira do D
´
efict de Atenc¸
˜
ao.
Miloud-Aouidate, A. and Baba-Ali, A. (2015). An effi-
cient ant colony instance selection algorithm for KNN
classification. International Journal of Applied Meta-
heuristic Computing, 4:47–64.
Moreira, S. C. and Barreto, M. A. M. (2017). Transtorno de
d
´
eficit de atenc¸
˜
ao e hiperatividade: conhecendo para
intervir. Revista Pr
´
axis, 1(2):65–70.
Muzetti, C. M. G. and de Luca Vinhas, M. C. Z. (2017).
Influ
ˆ
encia do d
´
eficit de atenc¸
˜
ao e hiperatividade na
aprendizagem em escolares. Psicologia argumento,
29(65):237–248.
Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V.,
Thirion, B., Grisel, O., Blondel, M., Prettenhofer,
P., Weiss, R., Dubourg, V., Vanderplas, J., Passos,
A., Cournapeau, D., Brucher, M., Perrot, M., and
Duchesnay, E. (2011). Scikit-learn: Machine learning
in Python. Journal of Machine Learning Research,
12:2825–2830.
Pyle, D. (1999). Data Preparation for Data Mining. Mor-
gan Kaufmann Publishers Inc., San Francisco, CA,
USA, 1st edition.
Rangel J
´
unior,
´
E. d. B. and Loos, H. (2011). Escola e
desenvolvimento psicossocial segundo percepc¸
˜
oes de
jovens com TDAH. Paid
´
eia, 21(50):373–382.
Retz, W., Ginsberg, Y., Turner, D., Barra, S., Retz-
Junginger, P., Larsson, H., and Asherson, P. (2020).
Attention-deficit/hyperactivity disorder (ADHD), an-
tisociality and delinquent behavior over the lifespan.
Neuroscience & Biobehavioral Reviews, 120:236–
248.
Salama, K. M., Abdelbar, A. M., and Anwar, I. M. (2016).
Data reduction for classification with ant colony algo-
rithms. Intelligent Data Analysis, 20:1021–1059. 5.
Santos, L. d. F. and Vasconcelos, L. A. (2010). Transtorno
do d
´
eficit de atenc¸
˜
ao e hiperatividade em crianc¸as:
uma revis
˜
ao interdisciplinar. Psicologia: Teoria e
Pesquisa, 26(4):717–724.
Selcuk, T. and Alkan, A. (2019). Detection of microa-
neurysms using ant colony algorithm in the early di-
agnosis of diabetic retinopathy. Medical hypotheses,
129:109242.
HEALTHINF 2022 - 15th International Conference on Health Informatics
110