Application of Formal Concept Analysis and Data Mining to

Characterize Infant Mortality in Two Regions of the State of Minas

Gerais

Deivid Santos, Cristiane Nobre, Luis Zarate and Mark Song

Instituto de Ci

encias Exatas e Inform

atica, Pontif

ıcia Universidade Cat

olica de Minas Gerais, Brazil

Keywords:

Formal Concept Analysis, Infant Mortality, SIM, SINASC.

Abstract:

Infant mortality is characterized by the death of children under one year, a problem that affects a large part of

the world population. This article applies the Formal Concept Analysis (FCA), a mathematical technique used

in data analysis to characterize infant mortality in two regions of Minas Gerais state - Brazil: Belo Horizonte

and Vale do Jequitinhonha. The Metropolitan Region of Belo Horizonte has the best human development rate,

and Vale do Jequitinhonha has the worst social equality. The relationships between attributes and victims are

identiﬁed through association rules and implications.

1 INTRODUCTION

Infant mortality is a term used to describe the death

of children during the ﬁrst year of life. It is a circum-

stance that occurs all over the world, more regularly

in poorer nations where one of the factors is low hu-

man development index and the lack of basic sanita-

tion, leading to an absence of healthy livelihoods and

the spreading of different diseases. The Infant Mor-

tality Rate is characterized as the number of deaths of

children under one year per thousand live births in a

speciﬁc population and year. As such, it is an indica-

tor that assesses the danger of a child conceived alive,

dying before completing the ﬁrst year of life.

Infant mortality can be understood as the sum

of deaths occurring in the early neonatal period (0

to 6 days of life), late neonatal (7 to 27 days), and

postneonatal period (28 days or more, up to a year)

(UNICEF, 2020).

The study of infant mortality can reveal insights

into which perspectives should be worked on in a pop-

ulation, with the aim that the extent of deaths can be

reduced, which is a deﬁnite factor for a country’s ad-

vancement. From a logical and social perspective, in-

fant mortality can be used as an approach to assess

networks and welfare strategies adopted in a deter-

mined region, as indicated by (Black et al., 2017). Be-

sides that, the newborn’s death studies can reveal per-

spectives around points of view, for example, the pop-

ulation’s diseases and the connection between social

disparity and infant mortality as indicated by (Her-

nandez et al., 2011). High estimates often reﬂect

doubtful degrees of well-being, bad health care, and

ﬁnancial changes.

Several factors can be associated with infant mor-

tality, as social and biological, presenting some dif-

ferences between regions with huge social imbalance.

According to data from (IBGE, 2021), the Brazilian

Institute of Geography and Statistics, in Vale do Je-

quitinhonha the infant death numbers are increasing

more than in other regions.

The federal government created the SIM (Mortal-

ity Information System), in 1975, in order to get data

about the children’s demise (Brazil, 2021a), and the

SINASC (System of Information about Born Alive),

in 1990, in order to get data about the children’s alive

(Brazil, 2021b). Both epidemiological reasonable-

ness structures acquired a reputation for presenting

the welfare, observing registers, and evaluating gov-

ernmental programs.

In this article, we characterized infant mortality

through SIM and SINASC repository data, applying

Formal Concept Analysis (FCA) in two regions of

Minas Gerais state: The Metropolitan Region of Belo

Horizonte (BH) with more than 6 million people, be-

ing the third largest in population brazil and the six-

tieth most populous metropolitan area in the world,

and the Vale do Jequitinhonha (VJ), bathed by the 31

thousand miles of the Jequitinhonha River, is home

to more than 950,000 Brazilians. The Vale Jequitin-

Santos, D., Nobre, C., Zarate, L. and Song, M.

Application of Formal Concept Analysis and Data Mining to Characterize Infant Mortality in Two Regions of the State of Minas Gerais.

DOI: 10.5220/0011039900003179

In Proceedings of the 24th International Conference on Enterprise Information Systems (ICEIS 2022) - Volume 1, pages 155-162

ISBN: 978-989-758-569-2; ISSN: 2184-4992

155

honha is divided into three micro-regions: Low, Mid-

dle, and High Jequitinhonha. The Lower Jequitin-

honha comprises the micro-region of Almenara, the

closest to the State of Bahia, while the Middle encom-

passes the regions of Pedra Azul and Arac¸ua

ı. Finally,

closer to the Metropolitan Region of Belo Horizonte,

there is the Alto Jequitinhonha. Minas Gerais is one

of the 27 Brazilian states and has 580 cities. The

Metropolitan Region of Belo Horizonte (BH) owns

35 of them and Vale Jequitinhonha (VJ), 51 (IBGE,

2021).

Formal Concepts Analysis (FCA) is a mathemat-

ical technique that had its beginnings in 1982 with

the work of Wille (Bernhard and Rudolf, 2012), who

considered each element of a reticulated to be a for-

mal concept, and the reticulated represented a hierar-

chy among the concepts. There is interest in applying

FCA in several areas such as health, software engi-

neering, data mining, among others. We use FCA in

this work to extract association and implication rules.

2 BACKGROUND

2.1 Formal Concept Analysis

The FCA is being used to recognize patterns with the

help of association rules and implications. The For-

mal Concept Analysis is composed of a set of ele-

ments such as formal context, formal concepts, and

rules. The formal context is a triple K(G, M, I), where

G and M are, respectively, a set of objects and a set of

attributes. I is an incidence relationship between G

and M, where I ⊆ G × M. Every incidence element is

denoted as gIm or (g, m) ∈ I.

A formal concept from a formal context

K(G, M, I) is deﬁned by a pair (A, B) where A is

the extention, and B the intention - A ⊆ G, B ⊆ M.

The pair (A, B) which deﬁnes the concept, need to

follow the conditions where (A = B’) e (B = A’).

This relation is deﬁned by the derivation operator as

follows:

= {g ∈ G | gIm ∀ m ∈ B} (1)

= {m ∈ M | gIm ∀ g ∈ G} (2)

The extension A contains each object from G that

has all the attributes from B, and the intention B owns

all the attributes from M that belongs to all objects in

Association rules are dependencies between ele-

ments of a formal context. A rule B → C is valid only

if every object that owns B attributes also contains C

attributes. Formally, B → C only if B, C ⊆ M and B’

⊆ C’. Given a rule r, and parameters s, c:

s = supp(r)=

∩B

is called the support of the rule r, and

c = conf(r)=

∩B

is its conﬁdence. When conf(r) = 100%, the rule is

denoted as an implication.

Developed por Rudolf Wille in the 80’s, the FCA

is a applicated mathematics camp based in the con-

cept and the conceptual hierarchy mathematization,

(Bernhard and Rudolf, 2012). The Formal Concept

Analysis considers the concepts as ways of intersub-

jectively comprehension in situations of oriented ac-

tion for the porpoise. The concepts formalization

must be clean and simple, but as well wide, so as the

main aspects of a concept can have its references ex-

plicit in the formal model.

2.1.1 Lattice Miner

Lattice Miner developed by the LARIM research lab-

oratory at Universit

e du Qu

ebec en Outaouais under

the supervision of Professor Rokia Missaoui. It al-

lows the generation of clusters (called formal con-

cepts) and association rules given a binary relation

between a collection of objects and a set of attributes.

The focus of Lattice Miner is on pattern (knowledge)

discovery visualization, exploration and approxima-

tion through a lattice representation of either a ﬂat or

a nested structure (Missaoui and Emamirad, 2017).

Version used in this work is (2.0).

3 RELATED WORKS

For countries in development, the child mortality until

the ﬁrst year still is a serious public health problem.

Although the child mortality in Brazil has reduces, it

continues to be high in different regions and cities.

In (Silva et al., 2017), they presents an FCA-based

approach to identify the behavior of professionals reg-

istered on linkedIn through the database. Appropriate

implications were applied to identify the skills that are

needed to achieve a particular job position. Applica-

tion of pre-processing techniques, transforming data

into a formal concept and ﬁnally extracting appropri-

ate implications using the PropIm algorithm analyz-

ing the results through graphical representations.

In (Barbosa et al., 2014) it was proposed char-

acterizing the 2008 child mortality in Vale Jequitin-

honha region analyzing the demographic and socioe-

conomic data, the mother’s historic, numbers of pre-

natal visits, data form a research done with the moth-

ers. The Epi Info for error identiﬁcation and posteri-

orly the Statistical Package for Social Science (SPSS)

ICEIS 2022 - 24th International Conference on Enterprise Information Systems

156

Figure 1: Methodology.

program for making statistics analysis. The study,

despite of the small data show, points that prenatal

period causes are determiners for child mortality, al-

though the social problems and difﬁculties consider-

ing health access has a huge interference as well. In

other (Barbosa et al., 2019) work it was proposed ana-

lyzing the child mortality focusing the evitability and

social vulnerability between 2009 and 2014, as well

in Vale Jequitinhonha region. The deaths were classi-

ﬁed and two groups: the evitable and inevitable ones,

following the CID-10 international disease classiﬁca-

tion. For the vulnerability analysis it was used the

IVS, the social vulnerability index. With the data

statistics analysis it was created a table that shows the

importance of the causes about biological factor from

the mother, and that evitable deaths are attributed as

an access problem and the lack of assistance by the

region health services.

In (Bonatti et al., 2020) it was proposed utilizing

polynomial regression models for the temporal series

study that analyzes the child mortality tendency in

Mato Grosso state, Brazil, between 2007 to 2016. The

result showed that regions had a child death reduction,

but maintained the same in a few other regions.

The Formal Concept Analysis use was utilized

(Ferreira et al., 2021) to extract a knowledge from a

base of 120 women who did breast cancer chemother-

apy, and they were separated in 2 groups to com-

pare antimedical or non-olanzapine treatments. The

Lattice Minner 2.0 was used for the rules extraction.

The results pointed capacity to help doctors and re-

searchers for proposals about the medicine adminis-

tration according to the patient response. In (Ananias

et al., 2021) it was proposed the classic FCA to the

Triadic Concept Analysis, proposing an approach to

ﬁnd triadic concepts, allowing exploration and visu-

alization, giving a tool for the data comprehension.

Utilized in (Noronha et al., 2020), the triadic analysis

allows the temporal evolution analysis from the clin-

ical conditions in order to identify proﬁles applying

the triadic rules that allows extracting human aging

information.

Although many papers present solutions regarding

this problem, most of them lack generalism. Some

related articles propose different methods. Others

address more speciﬁc issues by selecting a few at-

tributes. In some papers, it is impossible to identify

which procedures were adopted based on the infor-

mation provided. In this paper, we present and apply

a mathematical method of a binarized base for using

the FCA, which generates association and implication

rules, allowing us to determine infant mortality on as-

pects of mortality and survival, drawing a broad pic-

ture for each region.

4 METHODOLOGY

Although there are a lot of researches about the for-

mal concept analysis applied in the health area, just

a few explore the infant mortality rate characteriza-

tion in the Metropolitan Region of Belo Horizonte

and the Vale Jequitinhonha region, settled in Minas

Gerais, Brazil. This article investigates infant mor-

tality (IMR) using: data collection, analysis, explo-

ration, selection and attribution transformation, and

context and rules extraction.

The ﬁrst step was to collect data from the Unique

Health System Informatics Department (DATASUS),

unifying the databank. The strategic organ inside

the Brazilian government Health Ministry is respon-

sible for collecting, processing, and sharing informa-

tion related to health. Posteriorly, an analysis and at-

tributes exploration was made to understand and iden-

tify those non-relevant to the theme to reduce the di-

mensionality to work on the FCA approach.

In this article, we used the database from DATA-

SUS, considering the instances from 2006 to 2019

(for each region, DATASUS updates the information).

Application of Formal Concept Analysis and Data Mining to Characterize Infant Mortality in Two Regions of the State of Minas Gerais

157

The Unique Health System (SUS) has a more exten-

sive and variated database given by DATASUS. To ob-

tain the data, we used the Mortality Information Sys-

tem (SIM) (Brazil, 2021a), and the System of Infor-

mation about Born Alive (SINASC) to gather infor-

mation about infant mortality (Brazil, 2021b). Both

datasets acquired a reputation for spreading funda-

mental welfare information, observing registers, and

evaluating governmental programs. They were uni-

ﬁed so that a child classiﬁcation, that survived or not

their ﬁrst life year, could be done.

The database of infant mortality of the region of

Belo Horizonte taken from (DATASUS) and the Vale

Jequitinhonha region went through pre-processing

where irrelevant attributes, outliers were ﬁltered and

in the ﬁnal stage of database preparation was bal-

anced according to (Table 1), since an unbalanced

class leads to the classiﬁer being biased towards the

majority class (Prati et al., 2003).

Table 1: Balancing uneven datasets using undersampling.

class BH Region VJ Region

alive 1069565 → 1843 107858 → 87

dead 1843 → 1843 87 → 87

The attributes are numerical and categorical, con-

taining absent data and others with inconsistency

thatwere analyzed. The World Health Organization

(WHO) warns of the social and biological factors as-

sociated with infant mortality. They also use some of

these attributes in the works of (Soares et al., 2021)

and (Ananias et al., 2021). Based on this information,

selected the following attributes were: baby weight,

APGAR1, APGAR5, gestation, birth, type of preg-

nancy, congenital anomaly, marital status, mother’s

age, the number of sons and daughters alive, the num-

ber of sons and daughter dead, mother’s schooling,

race,and the number of antenatal visits (Sridevi and

Nirmala, 2016).

The attribute “weight” is a continuous numerical

that indicates the baby’s weight, in grams, when born.

For the data discretization, we used the DATASUS

tabulation (Brazil, 2017), which consists in weigh-

ing the baby at birth, in grams: superior or equal to

4000, between 3000 and 3999, 2500 and 2999, 1500

and 2499, 1000 and 1499, 500 and 999, and less them

500. It is possible to see the data discretization in as-

cending order (Figure 2). It shows that the percentage

of newborns in both the Belo Horizonte Metropoli-

tan Area and the Jequitinhonha Valley area is differ-

ent from 3000 to 3999 grams. We can also see dif-

ferences in the weights above 4000 grams and below

500 grams scales between 500 to 999 grams and 2500

to 2999 grams.

Figure 2: Weight of babies in the two regions evaluated (BH

region and VJ region).

The APGAR 1 (Figures 3) and APGAR 5 (Fig-

ures 4) are two measures widely used in the newborn

context evaluation. They are indicators used to evalu-

ate ﬁve objective signals of the newborn: (skin color,

reﬂective irritability, heart rate, breathing, and mus-

cular tonus), each item is assigned a score of 0 to 3,

the score of each item is added and a maximum score

of 10 points is obtained (Sykes et al., 1982).

The attribute “APGAR” scale is directly related to

mortality in the ﬁrst 28 days of the baby’s life. A

score between 8 and 10 indicates a normal condition

of the baby, a score between 4 and 7 the scale is con-

sidered moderate that needs special care, a score of 0

to 3 is severe and immediate resuscitation procedures

are required. The discretization was done based on

the apgar scale (Sykes et al., 1982). (Figure 3) shows

that between regions there are important differences

for both apgar1 considered good 8 to 10 and bad ap-

gar1 0 to 3. In (Figure 4) apgar5 on the scale of 0

to 3 shows equivalence in the data between the two

regions. Scales 4 to 7 and 8 to 10 show minor differ-

ences.

Figure 3: APGAR 1 value in the two regions evaluated (BH

region and VJ region).

The attribute “gestational” (Tabela 2) consists of

the time, in weeks, that the pregnancy lasted, was dis-

cretized used the DATASUS tabulation into six cate-

ICEIS 2022 - 24th International Conference on Enterprise Information Systems

158

Figure 4: APGAR 5 value in the two regions evaluated (BH

region and VJ region).

Table 2: Gestation period of mother in the two regions eval-

uated (BH region and VJ region).

Attribute BH region VJ region

42 weeks and more 1% 4%

37 to 41 weeks 54% 66%

32 to 36 weeks 13% 9%

28 to 31 weeks 9% 5%

22 to 27 weeks 22% 14%

less than 22 weeks 2% 2%

gories: less than 22 weeks, 22 to 27 weeks; 28 to 31

weeks; 32 to 36 weeks; 37 to 41 weeks and 42 weeks

and then grouped: preterm or premature pregnancies

( less than 37 weeks), the gestation of expected dura-

tion normal the gestational term is the period between

37 to 42 weeks, pregnancies exceeding 42 weeks are

called poster which happens less frequently. Accord-

ing to (WHO, 2021), attention to care is needed in

preterm and post-term babies.

Table 3: Type of Childbirth in the two regions evaluated

(BH region and VJ region).

Attribute BH region VJ region

normal 51% 69%

cesarean 49% 31%

The attribute “Child-birth” (Table 3) can be clas-

siﬁed as vaginal (or usual) or cesarean. While in the

Metropolitan Region Belo Horizonte the proportion

between vaginal and cesarean birth is balanced at al-

most 50%, in Vale do Jequitinhonha, the difference is

that the vaginal birth is 2 times more common than

the cesarean one.

Table 4: Type of pregnancy in the two regions evaluated

(BH region and VJ region).

Attribute BH region VJ region

only 93% 91%

twins 6% 9%

triple or more 1% 0%

The attributes “Birth type” indicates one of thefol-

lowing categories: only, twins, triple, and more babies

(Table 4).

Table 5: Congenital anomaly of babies in the two regions

evaluated (BH region and VJ region).

Attribute BH region VJ region

no anomaly 89% 91%

some anomaly 11% 9%

The attribute “Congenital anomaly” indicates that

the child had some anomaly. Congenital refers to the

existence at or before birth. It could occur during

pregnancy and could be detected before or after birth

(Table 5). Approximately 50% of all cannot be linked

to a speciﬁc disease, there are some known genetic,

environmental or risk factors.The great majority is no

anomaly birth.

Table 6: Marital status in the two regions evaluated (BH

region and VJ region).

Attribute BH region VJ region

single 52% 48%

married 43% 31%

consensual unions 3% 21%

divorced 2% 0%

The attribute “Marital status” identiﬁes the

mother’s social status divided into four categories:

single, married, divorced and consensual unions. As

it does not contain data in the widow category, this

attribute was excluded from the database (Table 6).

Figure 5: Age of mothers in the two regions evaluated (BH

region and VJ region).

The attribute “Mother’s age” (Figure 5) is a con-

tinuous numeral that indicates how old the baby’s

mother is, the discretization based on DATASUS tab-

ulation in eight categories: 10 to 14 years, 15 to 19

years, 20 to 24 years, 25 to 29 years, 30 to 34 years,

35 to 39 years, 40 to 44 years, and 45 to 49 years.

According to (WHO, 2021), adolescent mothers un-

der 20 years and mothers older than 35 years should

Application of Formal Concept Analysis and Data Mining to Characterize Infant Mortality in Two Regions of the State of Minas Gerais

159

receive special attention during pregnancy, for several

implications.

Table 7: Number Living Children (BH region and VJ re-

gion).

Attribute BH region VJ region

0 51,9% 40%

1 a 2 40,2% 43%

3 a 4 6,5% 8%

5 a 6 1% 7%

7 a 8 0,4% 2%

The attribute “Number of living children” (Table

7) was discretized according to the DATASUS tabula-

tion into ﬁve categories: 0 living children, between 1

and 2 children, 3 to 4 children, 5 to 6 children, and 7

to 8 living children. The number of children is linked

to Family Planning being a set of actions that help

men and women to plan the arrival of children, and

also to prevent unplanned pregnancy, thinking about

the conditions for their education and proper care.

Table 8: Number Dead Children (BH region and VJ region).

Attribute BH region VJ region

0 86% 85%

1 a 2 13% 13%

3 a 4 1% 2%

The attribute “number of dead children” (Table

8) were discretized, but due to lack of data, other

data ranges were excluded, leaving three categories:

0 dead children, 1 to 2 dead children, and 3 to 4 dead

children.

Table 9: Mother’s Schooling (BH region and VJ region).

Attribute BH region VJ region

1 to 3 years old 2% 13%

4 to 7 years old 18% 33%

8 to 11 years old 61% 48%

12 years and more 19% 6%

The attribute “Mother’s schooling” (Table 9),

shows some outliers, knowing that those instances

were removed. The attribute ”Mother’s schooling” in-

dicates the mother’s level of education in years. The

discretization was based on the DATASUS tabulation

(Brazil, 2017). The categories are 1 to 3 years, 4 to

7 years, 8 to 11 years, and 12 years and more. Ac-

cording (WHO, 2021), the categories below 8 years

of study are considered low education.

The attribute ”Race” (Table 10) has four cate-

gories: brown, white, black, and yellow. Therefore,

it is an essential indicator for analyzing the complex

social scenario according to race/color. The percent-

Table 10: Race categorization in the two regions evaluated

(BH region and VJ region).

Attribute BH region VJ region

brown 67% 83%

white 28% 12%

black 4% 4%

yellow 1% 1%

age of brown people in the Vale Jequitinhonha region

is higher if we consider the Belo Horizonte region,

which has a higher rate of white people.

All attributes were analyzed and modiﬁed accord-

ing to individual data analysis, discretized attributes

with missing or irrelevant data were excluded, gener-

ating the categorized base.

Table 11: Part of the formal context.

child-birth-normal

child-birth-cesarean

anomaly-not

anomaly-yes

weight-3000-3999

weight-2500-2999

weight-1500-2499

weight-1000-1499

weight-4000-more

weight-500-less-than

x x x

From the categorization, it was possible to create

a formal context by transforming the base with bina-

rized variables (Table 11).

We applied the Lattice Miner software, an FCA-

oriented tool, to generate the association and impli-

cation rules based on the formal context created. We

separated them into two scenarios, being living and

dead. We deﬁned a minimum support of 30% and a

conﬁdence level of 30%, to generate more rules. We

selected only some of the implication rules that had

the conﬁdence of 100%, higher support rate being the

consequence alive or dead. In (Table 12) and (Table

13) we selected for the two regions three main rules

for the baby survival scenario and three main rules for

the baby mortality scenario.

5 RESULTS AND DISCUSSION

When using the FCA, support and conﬁdence was de-

ﬁned with values of at least 30% for the scenario in

the metropolitan region of Belo Horizonte, the ﬁrst

analysis was considered the dead class, generating,

ICEIS 2022 - 24th International Conference on Enterprise Information Systems

160

this way, sixty-three (63) rules, later the Alive class

generated one hundred-sixty-nine (169) rules. For the

Jequitinhonha Valley region, the ﬁrst scenario gener-

ated one-hundred-ﬁfty-eight (158) rules for the dead

class, and for the alive class two hundred and eight-

hundred-sixty-six (866) rules.

It was possible to collect the results in xml format

and analyze the rules according to the form A → B, A

being the premise, and B the consequence, generating

the support (sup) and conﬁdence (conf) data. (Table

12) generates implication rules pointing attributes.

Rules number 6 and 5 indicate weight from 500 to

999 grams and a premature pregnancy around 22 to

27 weeks in the scenario for children who died. Rules

number 1, 2, and 3 indicate attributes as pregnancy

from 37 to 41 weeks, and antenatal visits are the most

explicit. In the scenario for living children, the most

reliable rule is 100%, and the most supportive is 86%.

Table 12: Main rules generated in Belo Horizonte Region.

N. Rules Sup Conf

IF antenatal visits 7,

gestation 37 to 41 weeks and

weight 3000 to 3999 grams

THEN Living

41% 100%

IF gestation 37 to 41 weeks

THEN Living

86% 100%

IF antenatal visits 7

THEN Living

72% 100%

IF apgar1 score of 0 to 3

THEN Child death

42% 100%

IF gestation 22 to 27 weeks

THEN Child death

43% 100%

IF gestation 22 to 27 weeks,

weight 500 to 999 grams

THEN Child death

33% 100%

In the region of Vale Jequitinhonha the rules in

the scenario for living children (Table 13) they point

different attributes as gestation time, only child, and

mother schooling. The rules selected were the 100%

conﬁdence ones and the rule with higher support is

85%.

Rules number 1, 2, and 3 point to gestation be-

tween 37 and 41 weeks and mother’s education above

8 years as crucial attributes for child survival. The

feature of a mother’s schooling does not appear for

the Metropolitan Region of Belo Horizonte, showing

it to be a decisive factor in the Vale Jequitinhonha Re-

gion.

For the rules in the scenario for living children

they point different attributes as gestation time, only

child, and mother schooling. The rules selected were

the 100% conﬁdence ones and the rule with higher

support is 85%.

Rule number 4 points out that even with an Apgar

score between 8 to 10, which is considered good, the

baby dies in 48% of cases. This is expected with an

Apgar score between 0 and 3, which, as rule 5 shows,

happens in 40% of the cases. Finally, rule number 6

points out the factor of the mother’s poor education as

one of the factors for the baby’s death.

Table 13: Main rules generated in Vale Jequitinhonha Re-

gion.

N. Rules Sup Conf

IF gestation 37 to 41 weeks

THEN Living

85% 100%

IF gestation 37 to 41 weeks

and only child

THEN Living

82% 100%

IF mother schooling 8 to 11

years old THEN Living

58% 100%

IF apgar1 score of 8 to 10

THEN Child death

48% 100%

IF apgar5 score of 0 to 3

THEN Child death

40% 100%

IF mother schooling 4 to 7

years oldTHEN Child death

48% 100%

In the Metropolitan Region of Belo Horizonte,

attributes such as weight, apgar, and gestation are

linked to the biological factors of the mother. Still,

with proper nutritional monitoring and adequate pre-

natal care, the three attributes above can be re-

duced and minimized since they are often among pre-

ventable causes. On the other hand, in the Vale Je-

quitinhonha region, the situation is a social issue, pre-

senting a low number of prenatal consultations due

to poor health structure and inadequate education of

some mothers, causing poor ﬁnancial conditions of

part of the population.

6 CONCLUSIONS

The association rules showed the characteristics of

infant mortality taking into account variables such

as weight and gestation time as determinants of

child survival, mortality is higher taking into account

Application of Formal Concept Analysis and Data Mining to Characterize Infant Mortality in Two Regions of the State of Minas Gerais

161

weight, premature pregnancy and APGAR with a

value between 0 and 3 , within the rules, it is remark-

able that for the metropolitan region of Belo Hori-

zonte, even if the child has an unfavorable weight,

the recovery is better than the region of Vale Je-

quitinhonha, which has higher mortality with higher

weights.

For future work, the necessity to create new sce-

narios and use different tools is essential to identify

and understand infant mortality in the regions of Mi-

nas Gerais and Brazil.

REFERENCES

Ananias, K. H., Missaoui, R., Ruas, P. H., Zarate, L. E., and

Song, M. A. (2021). Triadic concept approximation.

Information Sciences, 572:126–146.

Barbosa, T. A. G. d. S., Coelho, K. R., Andrade, G. N. d.,

Bittencourt, S. D. d. A., Leal, M. d. C., and Gazzinelli,

A. (2014). Determinantes da mortalidade infantil em

munic

ıpios do vale do jequitinhonha, minas gerais,

brasil. Revista Mineira de Enfermagem, 18(4):907–

922.

Barbosa, T. A. G. d. S., Gazzinelli, A., and Andrade, G.

N. d. (2019). Mortalidade infantil evit

avel e vulnera-

bilidade social no vale do jequitinhonha, minas gerais,

brasil. Revista Mineira de Enfermagem, 23:1–8.

Bernhard, G. and Rudolf, W. (2012). Formal concept anal-

ysis: mathematical foundations. Springer Science &

Business Media.

Black, R. E., Liu, L., Oza, S., Hogan, D., Perin, J., Rudan,

I., ELawn, J., Cousens, S., and Mathers, C. (2017).

Global, regional, and national causes of child mortal-

ityin 2000–13, with projections to inform post-2015

priorities: an updated systematic analysis. Lancet.

Bonatti, A. F., Silva, A. M. C. d., and Muraro, A. P. (2020).

Mortalidade infantil em mato grosso, brasil: tend

encia

entre 2007 e 2016 e causas de morte. Ci

encia & Sa

ude

Coletiva, 25:2821–2830.

Brazil (2017). Mortalidade geral – 1996

a 2015 notas t

ecnicas. website:

http://tabnet.datasus.gov.br/cgi/sim/Mortalidade Geral

1996 2012.pdf.

Brazil (2021a). Sistema informac¸

sobre mortalidade. website:

http://tabnet.datasus.gov.br/cgi/tabcgi.exe?sim/cnv/

obt10uf.def.

Brazil (2021b). Sistema informac¸

sobre nascidos vivos. website:

http://tabnet.datasus.gov.br/cgi/deftohtm.exe?sinasc/

cnv/nvuf.def.

Ferreira, L., Nobre, C., Z

arate, L., and Song, M. (2021).

Study of the evolution of antiemetic treatment through

the application of triadic formal concept analysis. In

Anais do IX Symposium on Knowledge Discovery,

Mining and Learning, pages 1–7. SBC.

Hernandez, A. R., Silva, C. H. d., Agranonik, M., Quadros,

F. M. d., and Goldani, M. Z. (2011). An

alise de

tend

encias das taxas de mortalidade infantil e de seus

fatores de risco na cidade de porto alegre, rio grande

do sul, brasil, no per

ıodo de 1996 a 2008. volume 27,

pages 2188–2196. SciELO Brasil.

IBGE (2021). Instituto brasileiro de geograﬁa e es-

tat

ıstica. website: https://www.ibge.gov.br/cidades-e-

estados/mg.html.

Missaoui, R. and Emamirad, K. (2017). Lattice miner-a

formal concept analysis tool. In 14th International

Conference on Formal Concept Analysis, page 91.

Noronha, M. D., Rodrigues, M. W., Ribeiro, C. E., Nobre,

C. N., Song, M. A., and Z

arate, L. E. (2020). Char-

acterization of long-lived and non-long lived proﬁles

through biclustering. In Proceedings of the 35th An-

nual ACM Symposium on Applied Computing, pages

473–476.

Prati, R. C., Batista, G., and Monard, M. C. (2003). Uma ex-

peri

encia no balanceamento artiﬁcial de conjuntos de

dados para aprendizado com classes desbalanceadas

utilizando an

alise roc. Proc. of the Workshop on Ad-

vances & Trends in AI for Problem Solving.

Silva, P. R. C., Dias, S. M., Brand

ao, W. C., Song, M. A.,

and Z

arate, L. E. (2017). Formal concept analysis

applied to professional social networks analysis. In

Proceedings of the 19th International Conference on

Enterprise Information Systems - Volume 1: ICEIS,,

pages 123–134. INSTICC, SciTePress.

Soares, W. L., Z

arate, L. E., Song, M. A., and Nobre, C. N.

(2021). Characterizing infant mortality using ma-

chine learningtechniques: a case study in two brazil-

ian states -santa catarina and amap

a. volume 7, pages

45269–45290.

Sridevi, S. and Nirmala, S. (2016). Anﬁs based decision

support system for prenatal detection of truncus arte-

riosus congenital heart defect. Applied Soft Comput-

ing.

Sykes, G., Johnson, P., Ashworth, F., Molloy, P., Gu, W.,

Stirrat, G., and Turnbull, A. (1982). Do apgar scores

indicate asphyxia? The Lancet, 319(8270):494–496.

UNICEF (2020). Neonatal mortality. website:

https://data.unicef.org/topic/child-survival/neonatal-

mortality/.

WHO (2021). World health organization. website:

https://www.who.int/data/gho/data/themes/topics/topic-

details/GHO/child-mortality-and-causes-of-death.

ICEIS 2022 - 24th International Conference on Enterprise Information Systems

162