Selection of Representative Instances Using Ant Colony Optimization:

A Case Study in a Database of Newborns with Congenital Zika in Brazil

Ana C. M. Gonc¸alves, Ludmila B. S. Nascimento

, Ana L. P. Leite, Maria E. O. Brito,

Erika G. de Assis, Henrique C. Freitas

and Cristiane N. Nobre

Department of Computer Science, Pontiﬁcal Catholic University of Minas Gerais,

Av. Dom Jos

e Gaspar, Belo Horizonte, Brazil

Keywords:

Zika, Ant Colony, Selection Instance, Microcephaly, Congenital Zika Syndrome.

Abstract:

This article investigates congenital syndrome associated with the Zika virus (ZIKV) in newborns in Brazil,

utilizing preprocessing techniques and machine learning to enhance its detection. The study proposes the Ant

Colony Optimization (ACO) algorithm for instance selection in a database on ZIKV infections from 2016,

during a period when Brazil faced a Zika outbreak linked to neurological complications such as microcephaly.

The research compares the performance of ACO with ﬁve classiﬁcation algorithms, demonstrating that ACO

improved all evaluation metrics. The highest case concentration was observed in Brazil’s Northeast and South-

east regions. Although cases have decreased in 2024, it is essential to maintain monitoring and preventive

actions. In summary, the results conﬁrm the effectiveness of ACO in enhancing machine learning models and

highlight the importance of clinical attributes in the early detection of congenital syndromes, recommending

the use of updated databases for a better understanding of the impact of ZIKV, particularly in newborns.

1 INTRODUCTION

Analysis, prediction, and decision-making for disease

diagnosis and treatment using data mining and ma-

chine learning require signiﬁcant effort. Existing al-

gorithms often need help to fully leverage large-scale

medical data and effectively analyze patient charac-

teristics (Carlin and Curran, 2012; Badawy et al.,

2023).

The growing volume of healthcare data, driven

by advancements in storage and collection, poses

challenges for data mining techniques due to redun-

dant or irrelevant attributes or instances. Attribute

and instance selection, as crucial preprocessing steps,

help mitigate this issue by eliminating data hinder-

ing learning performance and complicating modeling

(Akinyelu, 2020; Tsai et al., 2021).

The Ant Colony Optimization (ACO) algorithm,

inspired by the foraging behavior of actual ant

colonies, is widely used, for instance, and attribute

selection due to its efﬁciency in solving combinato-

https://orcid.org/0009-0004-9133-9671

https://orcid.org/0000-0001-9722-1093

https://orcid.org/0000-0001-8517-9852

rial problems (Anwar et al., 2015). ACO can ﬁnd

optimal or near-optimal solutions, making it a practi-

cal approach for reducing data sets while maintaining

classiﬁcation accuracy.

The Zika virus (ZIKV) is an arboviral disease

transmitted by the Aedes aegypti mosquito, ﬁrst iden-

tiﬁed in 1947 in Uganda, with human cases reported

since 1953 in Nigeria, according to the Minister of

Health

. The ﬁrst conﬁrmed case in the Americas

was in May 2015, in Northeast Brazil, and it rapidly

spread to other countries (Lowe et al., 2018). In 2016,

the WHO declared ZIKV a public health emergency

due to its association with congenital Zika syndrome

(Boeuf et al., 2016).

ZIKV can be transmitted from pregnant moth-

ers to fetuses, resulting in congenital anomalies such

as microcephaly and other neurological complica-

tions (Ribeiro and et al, 2018).

This study proposes using the RESP-

Microcephaly database, which documents ZIKV

cases in Brazil, focusing on newborns suspected of

having congenital syndrome. The aim is to apply

ACO, for instance, selection as a preprocessing

Available at: https://www.gov.br/saude/pt-

br/assuntos/saude-de-a-a-z/z/zika-virus

Gonçalves, A. C. M., Nascimento, L. B. S., Leite, A. L. P., Brito, M. E. O., G. de Assis, E., Freitas, H. C. and Nobre, C. N.

Selection of Representative Instances Using Ant Colony Optimization: A Case Study in a Database of Newborns with Congenital Zika in Brazil.

DOI: 10.5220/0013172600003911

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 18th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2025) - Volume 2: HEALTHINF, pages 587-594

ISBN: 978-989-758-731-3; ISSN: 2184-4305

587

technique to investigate the factors most inﬂuencing

the onset of congenital syndrome.

The analysis includes classiﬁcation algorithm

evaluation metrics such as F-Measure, Precision, and

Recall, comparing machine learning algorithms like

Decision Tree, AdaBoost, Random Forest, SVM, and

XGBoost.

2 BACKGROUND

2.1 Brazilian Geography

According to the Ministry of Foreign Affairs

Brazil

is the largest country in South America, covering

an area of 8,514,876 km² and hosting approximately

214 million inhabitants, distributed across ﬁve re-

gions: North, Northeast, Central-West, Southeast, and

South. These regions exhibit diverse climatic and ge-

ographic features, which signiﬁcantly inﬂuence the

transmission dynamics of mosquito-borne diseases

such as Zika virus, primarily transmitted by Aedes ae-

gypti.

The Northeast region, characterized by a humid

tropical climate along the coast and semi-arid con-

ditions inland, was the most affected during Brazil’s

Zika outbreak (Hartinger et al., 2023). High tempera-

tures, uneven rainfall, and water storage practices dur-

ing droughts created optimal breeding conditions for

mosquitoes. Similarly, with its dense urban popula-

tion and hot, rainy summers, the Southeast region ex-

perienced a signiﬁcant number of cases.

Climatic conditions such as prolonged heat, high

humidity, and rainfall patterns are critical to mosquito

life and disease spread. While the North region’s

equatorial climate supports year-round vector prolif-

eration, the Central-West and South regions exhibit

more seasonal risks. These ﬁndings emphasize the

need for region-speciﬁc public health interventions to

control the spread of the Zika virus (Hartinger et al.,

2023).

To better understand this subsection, Figure 1

presents the map of Brazil, highlighting its ﬁve re-

gions.

2.2 Zika Virus

The Zika virus (ZIKV) is an arbovirus transmitted

by arthropods belonging to the Flavivirus species

and the Flaviviridae family. In addition to ZIKV,

the Flavivirus species includes over 52 other viral

Available at: https://www.gov.br/mre/pt-

br/embaixada-bogota/o-brasil/geograﬁa

Source: https://encurtador.com.br/Yjisg

Figure 1: Map of Brazil Regions.

species, including dengue, yellow fever, Saint Louis

encephalitis, and West Nile viruses (Zanluca et al.,

2015). The Zika virus is primarily spread by the

vector Aedes aegypti, found in tropical and subtrop-

ical areas, and also by Aedes albopictuss, present in

the Mediterranean region of Europe (Carvalho et al.,

2019).

When a mosquito carrying the Zika virus bites

an individual, the insect’s saliva is injected into the

skin along with the virus. Components in this saliva

can exacerbate the viral infection by modifying the

immune system to favor cutaneous viral replication

(Hastings et al., 2019). After the bite, there is an in-

cubation period of approximately nine days, followed

by the onset of symptoms (Carvalho et al., 2019). Ac-

cording to the World Health Organization

, most peo-

ple infected with the Zika virus do not develop symp-

toms. When symptoms are present, they typically

include rash, fever, conjunctivitis, muscle and joint

pain, malaise, and headache, lasting between 2 to 7

days.

2.3 Newborns with Congenital Zika

According to Boeuf et al. (2016), ZIKV can infect

and damage neural progenitor cells, potentially im-

pacting fetal brain development and causing condi-

tions such as microcephaly and other neurodevelop-

mental anomalies.

ZIKV is the only vertically transmitted ﬂavivirus

that can infect cortical progenitor cells (Evans-

Gilbert, 2020). Transmission from mother to embryo

or fetus can occur during pregnancy or labor (Zanluca

et al., 2015).

ZIKV crosses the placental barrier, adversely af-

fecting embryonic development and triggering mi-

crocephaly through its impact on neural stem cells

Available at: https://www.who.int/news-room/fact-

sheets/detail/zika-virus

HEALTHINF 2025 - 18th International Conference on Health Informatics

588

(Evans-Gilbert, 2020).

Newborns suspected of microcephaly undergo

physical exams, head circumference measurements,

and neurological and imaging tests. Transfontanellar

ultrasound is the initial test of choice, with tomogra-

phy used when the fontanel is closed (Brasil et al.,

2015).

The Ministry of Health, following WHO recom-

mendations, adopted the InterGrowth-21st parame-

ters

for the ﬁrst 24-48 hours of life. According to

this reference, the head circumference of a 37-week

gestation child should be 30.24 cm for girls and 30.54

cm for boys. Accurate measurement, preferably to

two decimal places, is essential for proper assessment

(Brasil et al., 2015).

2.4 Instance Selection with Ant Colony

The Instance Selection (IS) technique aims to create

a subset of the original database by removing irrele-

vant, noisy, and redundant instances while maintain-

ing near-complete accuracy. This enhances data qual-

ity, reduces computational costs, and provides a min-

imal representative sample. Since IS involves search-

ing all possible combinations of instances, it is an

NP-Complete problem (Papadimitriou and Steiglitz,

1982) that requires heuristic solutions. This study em-

ploys the Ant Colony Optimization (ACO) heuristic

(Dorigo et al., 2006) to identify the best subset based

on the k-NN classiﬁer’s accuracy.

ACO is inspired by the behavior of ants ﬁnding the

shortest path using pheromones to guide their route

(Dorigo et al., 2006). In this study, the input instances

form the vertices of a graph, with Euclidean distances

deﬁning the edges. Artiﬁcial ants navigate this graph,

selecting components based on their heuristic value

and pheromone levels (Salama et al., 2016). Each ant

starts from a different instance and generates subsets

evaluated by machine learning algorithms. The best-

performing subset is retained as the solution.

The pheromone model F assigns parameters τ

i j

all paths, reﬂecting the colony’s accumulated knowl-

edge. High τ

i j

values indicate preferred paths. At

each iteration, pheromone values for used compo-

nents increase, while all values undergo evapora-

tion to guide future iterations towards better solutions

(Dorigo et al., 2006). This study uses the ANT-IS

method, proposed by Miloud-Aouidate and Baba-Ali

The INTERGROWTH-21st project developed a stan-

dard fetal growth curve for international use, aimed at

studying growth, health, nutrition, and neuromotor develop-

ment from 14 weeks of gestation to two years of age. This

standard complements the WHO growth curve for children

of both sexes

(2015), where an ant k at instance i at time t chooses

its next instance j based on Equations 1, 2, and 3.

1. Calculate the Euclidean distance between each in-

stance and all others in the dataset.

2. Initialize matrix C with -1 and pheromone values.

3. Place each ant on a unique instance.

4. Ants select their destination instance based on

Equations 1, 2, and 3.

5. Update the validation matrix C:

(a) If a

= 1, set C(k, j) = 1.

(b) Else, set C(k, j) = 0 and return to step 5.

6. Calculate path length L

(t) for each ant.

7. Compute and update pheromone values P

i j

(t) as

per Equations 3 and 4.

8. Repeat until all ants complete their tour; then pro-

ceed to step 11.

9. Keep the best solution with the highest classiﬁca-

tion rate from k-NN for k=1.

10. Clear the list of visited instances.

FProb

i j

= Prob

i j

∗ a

(1)

Prob

i j

(t) =











i j

(t) ∗ n

i j

,i f j ∈ N

∑

l∈N

(t)

∗ n

0,else

(2)

i j







∆P

i j

(t) =

(t)

,i f (i, j) ∈ T

(t)

0,else

(3)

Considering:

• T : set of instances

• n = |T |: number of instances

• b

(t): number of ants at instance i at time t

• n

i j

= 1/d

i j

: visibility of instance j for an ant at

instance i

• P

i j

: pheromone value on edge (i, j)

• C: matrix (b

(t)xn) containing the validation of the

destination points

• a

: random binary parameter

And, at the end of each iteration of the algorithm,

the pheromone value previously deposited on all paths

undergoes evaporation according to Equation 4.

i j

(t + 1) = (1 − ρ) ∗ P

i j

(t) +

∑

k=1

∆P

i j

(t) (4)

Selection of Representative Instances Using Ant Colony Optimization: A Case Study in a Database of Newborns with Congenital Zika in

Brazil

589

3 RELATED WORK

The section on related studies will explore the scope

and impact of Newborns affected by Congenital Syn-

drome resulting from Zika virus infection. It will

review recent research, intervention strategies, and

emerging public health issues related to this topic.

Additionally, studies using the Ant Colony Algo-

rithm, for instance selection, will be presented, along

with articles examining various aspects of the Zika

virus to enhance understanding of its spread and

global impact.

3.1 ZIKV

The article by Lowe et al. (2018) reviews the emer-

gence of Zika in Brazil, covering transmission routes,

clinical complications, and socioeconomic impacts. It

also identiﬁes knowledge gaps and challenges in pre-

venting future arbovirus outbreaks.

According to Zanluca et al. (2015), the rapid

spread of Zika in Brazil has reached over 50 coun-

tries in the Americas, with Aedes aegypti as the pri-

mary vector. The public health impact was signif-

icant, mainly due to neurological complications in

newborns, such as microcephaly. The swift spread

of Zika, compared to the slower spread of other ar-

boviruses like dengue, underscores the importance of

factors like climate and population mobility in shap-

ing outbreaks and highlights the need for effective

prevention measures.

3.2 Congenital Syndrome Caused by

the Zika Virus

A study by Ribeiro and et al (2018) investigated mi-

crocephaly cases in Piau

ı during the 2015–2016 Zika

epidemic. Researchers analyzed data from newborns

using the Live Births Information System (SINASC)

and medical records to assess maternal and infant

infections. Out of 75 microcephaly cases, 34 were

linked to congenital infections, with only one testing

positive for Zika IgM. Imaging tests conﬁrmed brain

anomalies in many cases.

In a review by Prata-Barbosa et al. (2019) on chil-

dren exposed to Zika during gestation, intrauterine

growth restriction and low birth weight were standard

among those with congenital Zika syndrome. Postna-

tal growth deﬁcits correlate with the severity of neu-

rological impairment, possibly inﬂuenced by nutri-

tional factors. The ﬁndings suggest that the impact

on growth in congenital Zika cases, whether or not

microcephaly was present at birth, is more signiﬁcant

with higher neurological impairment.

3.3 Instance Selection with Ant Colony

The preprocessing stage, for instance, selection is cru-

cial for enhancing the efﬁciency of machine learning

algorithms and data analysis. Many studies have ex-

plored instance selection heuristics, with Ant Colony

Optimization (ACO) recognized for its signiﬁcance in

this context.

Anwar et al. (2015) was among the ﬁrst to ap-

ply ACO, for instance, selection, extending the ADR-

Miner algorithm for data reduction to improve classi-

ﬁcation accuracy. This approach tested different clas-

siﬁcation algorithms at various stages of instance se-

lection to assess their effectiveness in building the ﬁ-

nal model.

Hott et al. (2022) demonstrated that ACO-based

instance selection improved the accuracy of classiﬁ-

cation models for identifying academic performance

in children and adolescents with ADHD, achieving

a 20 percentage point increase in K-NN accuracy.

This improvement shows the potential for early ed-

ucational intervention and more targeted support.

This section discusses two main themes: the im-

pacts of ZIKV and the application of ACO, for in-

stance, selection. It reviews studies on the rapid

spread of ZIKV, its neurological complications such

as microcephaly, and the challenges of controlling

outbreaks. Additionally, it highlights how ACO has

been effectively used to enhance predictive models,

such as identifying academic difﬁculties in students

and showcasing its beneﬁts for public health data

analysis.

Both topics underline the importance of pub-

lic health and advancements in computational tech-

niques. Analyzing large-scale public health data, like

Zika cases, with ACO can optimize instance selection

and uncover patterns to support more effective inter-

ventions.

4 MATERIALS AND METHODS

4.1 Dataset

The Ministry of Health provided the database used

in this research, which was collected through an on-

line form developed by DATASUS-Brasil and is avail-

able in the RESP-Microcefalia system. This system

aims to register suspected cases and deaths related to

growth and development alterations associated with

the Zika virus and other infectious diseases (Brasil

et al., 2015).

No variables identifying individuals or their fami-

lies were included, so submission for Research Ethics

HEALTHINF 2025 - 18th International Conference on Health Informatics

590

Committee review was not required, per the National

Health Council Resolution (CNS)

number 510, dated

April 7, 2016.

The RESP dataset includes 43 attributes organized

into nine categories, covering information about preg-

nant women, live births, pregnancy, delivery, and

more. These attributes are detailed in Table 1.

4.2 Preprocessing

The initial database contained 17,451 instances. To

focus on newborns with congenital anomalies linked

to Zika virus infection, only relevant instances were

selected, resulting in a ﬁnal dataset of 9,537 cases:

2,455 newborns diagnosed with congenital Zika syn-

drome and 7,082 without congenital anomalies.

Preprocessing steps were performed using Python

to prepare the data for classiﬁcation algorithms:

1) Missing Data Imputation: Addressed the 30%

of missing values by applying mean or median impu-

tation to minimize the impact of incomplete data.

2) One-Hot Encoding: Transformed ten nominal

categorical attributes into numerical values for com-

patibility with classiﬁcation algorithms.

3) Attribute Removal: Removed highly correlated

attributes, like brain diameter and head circumfer-

ence, to prevent biases. Retained head circumference

is a key criterion for microcephaly diagnosis. Non-

informative attributes, such as residential address and

phone number, were excluded.

4) Instance Selection: Used the Ant Colony Opti-

mization (ACO) algorithm to eliminate redundant and

noisy instances, reducing computational costs and

creating a more accurate sample. Details of the sub-

set generated by ACO are provided in the following

subsection.

4.3 The ACO’s Subset

The ACO algorithm was implemented and run on a

GPU using the NVIDIA CUDA library

to take ad-

vantage of its efﬁciency for faster and more effective

AI algorithm execution.

The ACO algorithm generated a subset of 4,866

instances, including 3,582 newborns without congen-

ital anomalies and 1,284 diagnosed with congenital

Zika syndrome. This subset was used in the experi-

ments described in the following sections.

Available at: https://abrir.link/pSSkm

https://developer.nvidia.com/cuda-toolkit

4.4 Evaluation Metrics

Five classiﬁcation algorithms were used: Decision

Tree, AdaBoost, Random Forest, Support Vector Ma-

chine (SVM), and XGBoost.

The Decision Tree was chosen for its simplicity

and interpretability. AdaBoost and Random Forest,

both ensemble tree methods, were selected for their

accuracy improvement and overﬁtting reduction capa-

bilities, respectively. SVM was included for its effec-

tiveness in high-dimensional spaces. XGBoost was

chosen for its high performance, speed, ability to han-

dle large datasets, and advanced overﬁtting prevention

techniques.

The dataset was split into 80% for training and

20% for testing. Ten-fold cross-validation was used

to assess model generalization

. The entire pipeline,

including model training and graph generation, was

executed in Python 3.10.9.

The ACO and machine learning algorithms ran on

a system with an Intel® Xeon® E5-2696 v3 Proces-

sor, 128 GB of DDR4 RAM (2400 MHz), and Ubuntu

Server 22.04 LTS.

To assess the ACO’s effectiveness in selecting op-

timal instances, three evaluation metrics were used:

F-measure

, Precision

, and Recall

5 RESULTS AND DISCUSSIONS

This section presents the results of analyzing of the

machine learning models obtained by the Ant Colony

Optimization (ACO) algorithm on the “Newborns

with Congenital Zika” database.

5.1 Results of the Machine Learning

Models

Figure 2 compares the metric values for the machine

learning algorithms obtained before and after apply-

ing instance selection, covering the ﬁve classiﬁcation

algorithms used.

Overall, AdaBoost, Decision Tree, Random For-

est, and XGBoost produced similar results in both ver-

sions. In contrast, SVM showed lower performance.

Using ACO, for instance, selection improved all three

metrics: the F-Measure increased by 5%, precision by

Cross-validation divides the data into training and test-

ing sets, evaluating the model across multiple parts to avoid

performance bias.

F-measure=

2×Precision×Recall

Precision+Recall

Precision =

TruePositive

TruePositive+FalsePositive

Recall =

TruePositive

TruePositive+FalseNegative

Selection of Representative Instances Using Ant Colony Optimization: A Case Study in a Database of Newborns with Congenital Zika in

Brazil

591

Table 1: Division of Attributes in the RESP Dataset.

Category Attributes

Notiﬁcation

Classiﬁcation of suspected cases of congenital infection

Date it was notiﬁed

Data on pregnant women

Age

Race/Color

State of residence

Information about live births

Sex

Date of birth

Weight

Height

Data about pregnancy and childbirth

Types of congenital changes

Timing of alteration detection

Gestational age of detection of microcephaly

Type of pregnancy

Live birth classiﬁcation

Head circumference

Date of head circumference measurement

Clinical and epidemiological data of the mother

Date of symptom onset

Types of symptoms

STORCH test conduction and result

Zika test result

History of arboviruses

Congenital malformation

Information on imaging exams

Ultrasound

Transfontanellar ultrasound

Computed tomography

Magnetic resonance imaging

Data on healthcare establishment

City

State

Data on disease progression

Death

Date of death

Restricted access ﬁelds for the manager

Final classiﬁcation of the suspected case of congenital alterations

Conﬁrmation criteria through laboratory tests performed

Figure 2: Performance of Machine Learning Algorithms.

7%, and recall by 5%. In conclusion, the dataset with

ACO selection outperformed the original in machine

learning applications.

5.2 Analysis of the Key Determinant

Attributes for Zika Classiﬁcation

After applying machine learning algorithms, the most

important attributes for identifying whether a new-

born had congenital Zika syndrome were extracted

from the ACO-reduced dataset. The top seven at-

tributes identiﬁed were: 1) Imaging Examinations; 2)

Head Circumference; 3) Newborn Weight; 4) Pres-

ence of Rash; 5) Zika Test Result; 6) Mother’s Age; 7)

Newborn Length.

According to Brasil et al. (2015, 2017), imaging

examinations are essential for detecting neurological

anomalies such as microcephaly, often linked to con-

genital infections like the Zika virus. Among new-

borns diagnosed with congenital Zika, 1,008 under-

went imaging examinations, accounting for 78%.

These ﬁndings align with previous research

(Brasil et al., 2017) highlighting newborn weight,

length, and head circumference as key diagnostic

parameters for congenital abnormalities, particularly

microcephaly, within the Intergrowth growth curve.

As noted by Brasil et al. (2017), exanthema (rash)

is a common symptom in pregnant women infected

with Zika, often accompanied by fever, joint pain,

conjunctivitis, and itching. This indicator can be asso-

ciated with fetal complications, including congenital

malformations. In the study, 634 newborns had moth-

ers who experienced exanthema during pregnancy,

representing 49%.

Maternal variables, such as age, were also sig-

niﬁcantly associated with congenital malformations

(Brasil et al., 2017).

Only 318 newborns diagnosed with congenital

HEALTHINF 2025 - 18th International Conference on Health Informatics

592

Zika underwent laboratory testing, making up just

25% of the total. This suggests that other factors be-

yond lab results play a crucial role in diagnosis, high-

lighting the need for a comprehensive diagnostic ap-

proach.

The ACO-generated dataset also revealed the dis-

tribution of congenital Zika cases by region in Brazil

during the 2016 outbreak, shown in Figure 3. The

Northeast region had the highest number of cases, fol-

lowed by the Southeast, indicating a signiﬁcant re-

gional impact.

Figure 3: Conﬁrmed Zika cases, separated by Brazilian re-

gion

The Northeast region of Brazil, with its hu-

mid tropical climate along the coast and semi-arid

conditions inland, provides favorable conditions for

mosquito proliferation, as noted by Hartinger et al.

(2023). This region reported the ﬁrst conﬁrmed Zika

case in May 2015 Lowe et al. (2018) and has the high-

est number of newborns diagnosed with congenital

Zika.

The Southeast region, Brazil’s most populous

area, with over 90% of its population in urban set-

tings, also has conditions conducive to mosquito-

borne disease spread due to hot, rainy summers. This

explains why it ranks second for congenital Zika

cases, especially during the rainy season when cli-

matic conditions favor the virus’s spread.

Although this study used the 2016 RESP dataset

from the Zika outbreak, Zika remains a public health

concern. The Minist

erio’s Epidemiological Report

No. 07

reported 243,720 dengue cases from the ﬁrst

to the fourth week of January 2024, a 273% increase

compared to 2023. Zika cases during the ﬁrst to third

weeks of January 2024 totaled 105, showing a 63%

decrease from 2023. While this decrease is encourag-

ing amid rising dengue cases, continuous monitoring

Available at: https://abrir.link/wiebX

is essential.

Analysis of the Ministry of Health’s arbovirus up-

date panel

shows that Zika cases were similar in

February 2023 (1,013 cases) and 2024 (1,010 cases).

This stability suggests the need for ongoing, adap-

tive efforts to maintain effective control and prevent a

resurgence, ensuring public health remains protected.

6 FINAL CONSIDERATIONS

The results conﬁrm the effectiveness of instance se-

lection using the Ant Colony Optimization (ACO)

algorithm in enhancing the performance of machine

learning models on the “Newborns with Congenital

Zika” dataset. ACO improved Precision, Recall, and

F-Measure metrics by 7%, 5%, and 5% compared to

the version without instance selection.

These improvements demonstrate that focusing on

relevant attributes can optimize the classiﬁcation pro-

cess and enhance accuracy. In healthcare, ACO in-

stance selection proved valuable for improving diag-

nostic precision, aiding early detection of congenital

syndromes, and supporting better medical decision-

making and resource allocation.

Key diagnostic attributes for Congenital Zika,

such as head circumference, weight, and exanthema,

showed strong correlations with previously identiﬁed

clinical characteristics. Maternal age and imaging test

results further reinforced the importance of variables

for early diagnosis and identiﬁcation of congenital

complications.

Geographical analysis showed the highest concen-

tration of congenital Zika cases in Brazil’s Northeast

and Southeast regions, highlighting their vulnerability

to Zika epidemics, especially under conditions favor-

able to mosquito proliferation.

The decline in Zika cases in 2024, as noted in

Ministry of Health data, is a positive trend, but on-

going monitoring and prevention remain critical. The

disease’s persistence at stable levels underscores the

need for continued and improved control measures

to protect public health against arbovirus threats in

Brazil.

Future research should incorporate updated

datasets on Congenital Zika cases to provide a com-

prehensive view of the disease’s current impact. Com-

paring 2016 data with more recent information will

help identify changes in epidemiological patterns and

enhance public health responses. This approach will

strengthen government and health service efforts to

Available at: https://abrir.link/Uxqat

Selection of Representative Instances Using Ant Colony Optimization: A Case Study in a Database of Newborns with Congenital Zika in

Brazil

593

combat Zika and its consequences, ensuring interven-

tions are based on current and accurate data.

ACKNOWLEDGMENTS

The authors would like to thank the National Coun-

cil for Scientiﬁc and Technological Development of

Brazil (CNPq – Code: 311573/2022-3), the Co-

ordination for the Improvement of Higher Educa-

tion Personnel - Brazil (CAPES - Grant PROAP

88887.842889/2023-00 - PUC/MG, Grant PDPG

88887.708960/2022-00 - PUC/MG - Informatics and

Finance Code 001), the Foundation for Research

Support of Minas Gerais State (FAPEMIG – Codes:

APQ-03076-18 and APQ-05058-23).

REFERENCES

Akinyelu, A. A. (2020). Bio-inspired technique for improv-

ing machine learning speed and big data processing.

In 2020 International Joint Conference on Neural Net-

works (IJCNN), pages 1–8.

Anwar, I. M., Salama, K. M., and Abdelbar, A. M. (2015).

Instance selection with ant colony optimization. Pro-

cedia Computer Science.

Badawy, M., Ramadan, N., and Hefny, H. (2023). Health-

care predictive analytics using machine learning and

deep learning techniques: a survey. J. of Electric. Syst.

and Inform. Techno., 10.

Boeuf, P., Drummer, H., Richards, J., Scoullar, M., and

Beeson, J. (2016). The global threat of zika virus

to pregnancy: Epidemiology, clinical perspectives,

mechanisms, and impact. BMC Medicine, 14(1).

Brasil, da Sa

ude, M., and de Vigil

ancia em Sa

ude, S.

(2015). Protocolo de vigil

ancia e resposta

a ocorr

encia

de microcefalia e/ou alterac¸

oes do sistema nervoso

central (snc):.

Brasil, da Sa

ude, M., and de Vigil

ancia em Sa

ude, S.

(2017). Orientac¸

oes integradas de vigil

ancia e atenc¸

a sa

ude no

ambito da emerg

encia de sa

ude p

ublica de

import

ancia nacional.

Carlin, S. and Curran, K. (2012). Cloud computing tech-

nologies. International Journal of Cloud Computing

and Services Science (IJ-CLOSER), 1.

Carvalho, I., Alencar, P., Andrade, M., Silva, P., Carvalho,

E., Ara

ujo, L., Cavalcante, M., and Sousa, F. (2019).

Clinical and x-ray oral evaluation in patients with con-

genital Zika Virus. Journal of applied oral science :

revista FOB, 27:e20180276–e20180276.

Dorigo, M., Birattari, M., and Stutzle, T. (2006). Ant colony

optimization. IEEE computational intelligence maga-

zine, 1(4):28–39.

Evans-Gilbert, T. (2020). Vertically transmitted chikun-

gunya, zika and dengue virus infections: The patho-

genesis from mother to fetus and the implications

of co-infections and vaccine development. Interna-

tional Journal of Pediatrics and Adolescent Medicine,

7(3):107–111.

Hartinger, S. M., Yglesias-Gonz

alez, M., Blanco-

Villafuerte, L., Palmeiro-Silva, Y. K., Lescano, A. G.,

Stewart-Ibarra, A., Rojas-Rueda, D., Melo, O., Taka-

hashi, B., Buss, D., Callaghan, M., Chesini, F., Flores,

E. C., Posse, C. G., Gouveia, N., Jankin, S., Miranda-

Chacon, Z., Mohajeri, N., Helo, J., and Ortiz, L.

(2023). The 2022 south america report of the lancet

countdown on health and climate change: trust the sci-

ence. now that we know, we must act. The Lancet

Regional Health - Americas, 20:100470–100470.

Hastings, A. K., Hastings, K., Uraki, R., Hwang, J.,

Gaitsch, H., Dhaliwal, K., Williamson, E., and Fikrig,

E. (2019). Loss of the TAM Receptor Axl Ameliorates

Severe Zika Virus Pathogenesis and Reduces Apopto-

sis in Microglia. iScience, 13:339–350.

Hott, H. A. C. F., Jandre, C., Xavier, P., Miloud-Aouidate,

A., Miranda, D., Song, M., Z

arate, L. E., and Nobre,

C. (2022). Selection of representative instances using

ant colony: A case study in a database of children and

adolescents with attention-deﬁcit/hyperactivity disor-

der. In International Conference on Health Informat-

ics.

Lowe, R., Barcellos, C., Brasil, P., Cruz, O., Honorio, N.,

Kuper, H., and Carvalho, M. (2018). The zika virus

epidemic in brazil: From discovery to future impli-

cations. International Journal of Environmental Re-

search and Public Health, 15:96.

Miloud-Aouidate, A. and Baba-Ali, A. (2015). An efﬁcient

ant colony instance selection algorithm for knn classi-

ﬁcation. International Journal of Applied Metaheuris-

tic Computing, 4:47–64.

Papadimitriou, C. H. and Steiglitz, K. (1982). Com-

binatorial optimization: algorithms and complexity.

Prentice-Hall, Inc., USA.

Prata-Barbosa, A. M., Melo, M., Guastavino, A. B., and

Cunha, A. J. L. A. d. (2019). Effects of Zika infection

on growth,. Jornal de Pediatria, 95:S30 – S41.

Ribeiro, I. G. and et al (2018). Microcefalia no Piau

ı, Brasil:

estudo descritivo durante a epidemia do v

ırus Zika,

2015-2016. Epidemiologia e Servic¸o de Sa

ude.

Salama, K. M., Abdelbar, A. M., and Anwar, I. M. (2016).

Data reduction for classiﬁcation with ant colony algo-

rithms. Intelligent data analysis, 20(5):1021–1059.

Tsai, C.-F., Sue, K.-L., Hu, Y.-H., and Chiu, A. (2021).

Combining feature selection, instance selection, and

ensemble classiﬁcation techniques for improved ﬁ-

nancial distress prediction. J. Bus. Res.

Zanluca, C., Melo, V., Mosimann, A., Santos, G., Santos,

C., and Luz, K. (2015). First report of autochthonous

transmission of zika virus in brazil. Memorias do In-

stituto Oswaldo Cruz, 110.

HEALTHINF 2025 - 18th International Conference on Health Informatics

594