Application of Formal Concept Analysis to Characterize Driving

Behaviors and Socio-Cultural Factors Related to Driving

Diogo Miranda, Luis Z

arate and Mark Song

Instituto de Ciencias Exatas e Inform

atica, Pontif

ıcia Universidade Catolica de Minas Gerais, Brazil

Keywords:

Formal Concept Analysis, Aggressive Driving Behaviors, Safe Driving Behaviors, Mindsponge Theory.

Abstract:

This article addresses the global concern for road safety, where frequent accidents on roads and streets result in

loss of human lives, severe injuries, and signiﬁcant material damage, impacting not only the direct victims but

also their families and society at large. To tackle this challenge, it is crucial to analyze the factors contributing

to these accidents, particularly driver behaviors. This study investigates reckless behaviors such as speeding,

less obvious inﬂuences such as personality traits and sociocultural factors. Using Formal Concept Analysis

(FCA), the research examines a database containing information about Chinese drivers, aiming to provide

valuable insights for accident prevention and the promotion of safer road behaviors. In summary, the article

aims to deepen the understanding of factors related to trafﬁc accidents with the goal of enhancing road and

street safety.

1 INTRODUCTION

Trafﬁc accidents are a worldwide concern, claiming

lives, causing severe injuries, and resulting in signif-

icant material losses. These incidents occur daily on

the roads and streets of all nations and are one of the

leading causes of death. Their consequences impact

not only the direct victims but also their families and

society as a whole. To comprehend and mitigate this

problem, it is crucial to analyze and understand the

factors that contribute to the occurrence of these acci-

dents.

Trafﬁc accidents result in approximately 1.3 mil-

lion fatalities every year (approximately 3,000 per

day), leaving both men and women injured in non-

fatal accidents around the world. (World Health Or-

ganization, 2023)

The study of trafﬁc behaviors can provide insights

into areas that need to be addressed from a perspective

aimed at identifying behaviors and factors that lead to

these accidents, with the goal of reducing the number

of fatalities. This is a crucial aspect of a country’s

development process. Furthermore, studies on trafﬁc

behaviors can reveal non-trivial aspects that are chal-

lenging to identify, such as the effects of personality

combined with socio-cultural characteristics, result-

ing in risks and contributing to these accidents, as in-

dicated in (Yang et al., 2013).

There is a factor analyzed in trafﬁc accidents that

contributes to the cause of accidents, and that is reck-

less driving. This includes speeding, dangerous over-

taking, disregarding trafﬁc laws, and the use of elec-

tronic devices while driving. In addition to these fac-

tors, socio-cultural and educational aspects have a sig-

niﬁcant impact on road safety, as was analyzed in

(Houston et al., 2003).

In this article, the aim is to analyze the socio-

cultural aspects of these drivers, along with external

inﬂuences from friends and/or family, applying For-

mal Concept Analysis (FCA) to the database contain-

ing information about Chinese drivers (published in

August 2023), following a data processing available

in (Jin et al., 2023).

2 BACKGROUND

2.1 Formal Concept Analysis

Formal Concept Analysis (FCA) can be used to rec-

ognize patterns with the help of association rules and

their implications. Formal Concept Analysis consists

of a set of objects in a formal context, formal con-

cepts, and rules. A formal context can be represented

as a triple K = (G, M, I), consisting of a set of G ob-

jects, a set of M attributes, and an incidence relation

I ⊆ G × M with (g, m) ∈ I meaning that object g has

Miranda, D., Zárate, L. and Song, M.

Application of Formal Concept Analysis to Characterize Driving Behaviors and Socio-Cultural Factors Related to Driving.

DOI: 10.5220/0012335000003657

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

In Proceedings of the 17th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2024) - Volume 2, pages 56-62

ISBN: 978-989-758-688-0; ISSN: 2184-4305

attribute m.”. For a set of objects A ⊆ G, the set of

common attributes for the objects. of A is denoted by.

A :=

{

m ∈ M|∀g ∈ A : (g, m) ∈ I

}

, similarly, the set of

ommon attributes for the objects of B is denoted by

B :=

{

g ∈ G|∀m ∈ B : (g, m) ∈ I

}

A formal concept of a formal context K =

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

extension and B is called intention. For a pair (A, B)

to be considered a concept, one needs to follow the

condition where. (A = B

′

) and (B = A

′

). The set of

formal concepts of a context K is said to be β (K).

Association rules are dependencies between elements

of a formal context.

The rule A → B is valid only if for every object

containing attributes B, it also contains attributes from

C. Given a rule r and parameters s and c, one can

denote:

s = suppr (r) =

′

∩B

′

- called the support of rule

r, and

c = con f (r) =

′

∩B

′

- called conﬁdence.

When con f (r) = 100% the rule is referred to as

an implication. (Felde and Stumme, 2023)

Studies that explore domains that can be repre-

sented as a binary tabular base of objects and at-

tributes can often apply FCA. Longitudinal study ap-

proaches aim to investigate a sample of individuals

with certain characteristics over consecutive time pe-

riods, referred to as waves. On the other hand, FCA

is an approach in formal set theory that focuses on the

representation and analysis of the semantic structure

of data at a single point in time, without considering

evolution or changes over time.

FCA is based on the idea that concepts can be de-

ﬁned based on the relationships between objects and

attributes, enabling the creation of conceptual hierar-

chies and the understanding of associations between

meaningful terms. It is a useful technique for orga-

nizing and extracting information from data sets.

The dimensionality of a database is a crucial point

when attempting to generalize and ﬁnd relationships

within data. A low-dimensional database is one that

contains few samples from a speciﬁc domain. For

example, databases related to human behavior often

have low dimensionality when we want to analyze be-

havior within a certain population. One application

of FCA is to understand how the objects present in

these low-dimensional databases, along with their at-

tributes, can have implications.

2.2 Aggressive Behaviors in Trafﬁc

Behaviors in trafﬁc that can lead to accidents are char-

acterized into three main categories: 1) aggressive be-

haviors in trafﬁc, 2) inﬂuence of friends and close ac-

quaintances, 3) family inﬂuence. These categories en-

compass a fourth one that is used as a threshold for

analysis, which is socio-cultural information about

the drivers. The information pertains to 1039 Chi-

nese drivers whose sociocultural factors were asso-

ciated with these behaviors. The database was col-

lected through an online survey, publicly available in

the Data in Brief journal. This study was published

in August 2023 and has the Bayesian Mindsponge

Framework (BMF) as a validation index, speciﬁcally

showing how safe driving behaviors are affected by

information that promotes safe driving, actively ab-

sorbed with the support of friends/colleagues and/or

the driver’s family.

A fundamental concept of the Mindsponge Theory

is that the human mind tends to be inﬂuenced by infor-

mation absorbed from external sources. As analyzed

in (Jin et al., 2023), the factors that contribute to safe

driving may be related to external factors from family,

friends, and/or colleagues. These factors, along with

socio-cultural factors, provide interesting information

to be analyzed in this issue.

The application of FCA to the dataset in question

can provide important insights into the behavior of

drivers in trafﬁc leading to accidents. Rules of the

form A → B take into account that when drivers ex-

hibit a certain aggressive behavior A, it implies B.

2.3 Lattice Miner

The Lattice Miner 2.0 tool is a data mining prototype

developed under the supervision of Professor Rokia

Missaoui by the laboratory of the University of Que-

bec. This is a publicly available Java platform in

which the main functions include all low-level opera-

tions that allow the manipulation of input data, struc-

tures, and rule association. The platform enables the

generation of groups, called formal concepts, includ-

ing logical implications, thereby showing binary rela-

tionships between collections of objects and their sets

of attributes or properties.

3 RELATED WORKS

This work utilizes Formal Concept Analysis, and the

approach is justiﬁed through a relationship between

theoretical and practical knowledge of this subject.

Related works on this topic are presented below.

(Wei et al., 2018) analyzes the triadic approach of

formal concept analysis in four aspects: (i) the basic

approach of triadic concept analysis, (ii) triadic im-

plications and rules, (iii) the triadic factor of analysis,

Application of Formal Concept Analysis to Characterize Driving Behaviors and Socio-Cultural Factors Related to Driving

and (iv) the analysis of fuzzy triadic concepts.

(Biedermann, 1997) systematically demonstrates

the application of triadic formal concept analysis in

databases to represent complex concepts that are dif-

ﬁcult to visualize. It also explains the generation of

rules and implications from an analyzed dataset.

The work (Ganter and Obiedkov, 2004) demon-

strates various biases that can be generated from tri-

adic formal concept analysis and their implications

in multiple scenarios. Given different interests that

can be addressed from the triadic context, the authors

provide extensive and concise descriptions through

implication-generating algorithms in the triadic con-

text. Examples of these interests from various do-

mains, but still addressable in a triadic context, are

found in (Kent and Neuss, 1997), where the focus is

on hypertext analysis, and in (Carullo et al., 2015),

which presents an approach of this method in online

recommendation systems.

In (Blevente Lorand Kis and Troanca, 2017), a

tool is presented that enables the visualization of these

concepts and rules, facilitating navigation and under-

standing of triadic concepts.

(Hu et al., 2004) presents modeling techniques

based on a logical description language in a can-

cer database. Results are presented that are gener-

ated from intentions and extensions of entities present

in these databases, obtained through formal concept

analysis.

In the work (Deivid Santos, 2022), an analysis of

infant mortality in two regions of Minas Gerais is con-

ducted. The process used the approach of triadic for-

mal concept analysis to extract rules and implications

from a database. The study generated a series of rules

with certain hierarchies characterizing this database.

In (Lucas Ferreira, 2021), an application of a

process to extract knowledge from a database gen-

erated by a study conducted on women undergo-

ing chemotherapy treatment for breast cancer is per-

formed. The application of the Formal Concept Anal-

ysis theory allowed for the extraction of a set of hi-

erarchically organized concepts, from which rules re-

lating them were extracted, thus describing the out-

comes of antiemetic treatments in this database.

In (Paulo Lana, 2022), a longitudinal analysis of

a Covid-19 database is conducted using the processes

of triadic formal concept analysis. The results of this

work provide implication rules that longitudinally de-

scribe the evolution of the Covid-19 pandemic at dif-

ferent time points.

4 METHODOLOGY

There are many articles that deal with formal concept

analysis applied to the ﬁeld of health and human be-

havior. This speciﬁc article aims to investigate ag-

gressive behaviors of Chinese drivers in trafﬁc, using

data collection, exploration, attribute selection, and

transformation, as well as the extraction of contexts

and rules (Figure 1).

Figure 1: Metodology.

4.1 Materials

The dyadic database used in this study is available

in (Jin et al., 2023). The database contains records

of 1039 Chinese drivers who responded to a ques-

tionnaire. The questionnaire consists of 37 variables,

including responses from different perspectives. Ta-

ble 1 shows an example of the attributes extracted

from these questionnaires, with only the ﬁrst three

attributes from each category displayed for simpliﬁ-

cation purposes. The study was divided into groups,

each about the analyzed subcategories. Each subcate-

gory has a speciﬁc number of attributes.

4.2 Methods

4.2.1 Preprocessing

The ﬁrst step will be to collect the necessary data

and its description. Validated data is available in the

Data in Brief journal data repository (Jin et al., 2023).

The available data contains objects and attributes re-

lated to Chinese drivers, where attribute types can be

categorical and numerical. The database has been

pre-processed, with irrelevant attributes removed, and

outliers ﬁltered. Data balancing was not performed,

HEALTHINF 2024 - 17th International Conference on Health Informatics

Table 1: Variables available in the database.

Category Variable Question

Driving and insurance purchasing information

A1 Commercial insurance for vehicle

A2 Frequency of driving for work

A3 Conﬁdence in driving skills

Aggressive driving behaviors

B1 Speed limit: rarely exceed

B2 Normal speed, avoid weaving, reckless overtaking

B3 Safe distance, no tailgating

Friend/Peer Inﬂuence

C1 Supportive friends

C2 Advocating safe driving

C3 Caution against driving under the inﬂuence

Family Inﬂuence

D1 Planned driving

D2 Follow trafﬁc rules

D3 Praise safe driving

Socio-demographic factors

E1 Gender

E2 Education level

E3 Monthly salary

as all data in the database corresponds to the class that

the study seeks to characterize.

It was necessary to create categories for the anal-

ysis of the objects at hand. To do this, sub-categories

were created, and the selection of which attributes

would be present in them was made. Initially, the

analysis to be conducted is about the inﬂuence of

external factors on the behavior of drivers in trafﬁc.

For this purpose, a Cartesian product of general cat-

egories into sub-categories was performed. The at-

tributes in the categories related to driving and insur-

ance purchase information (A), socio-demographic

factors (E), and aggressive behaviors (B) were con-

sidered general attributes, meaning they are present in

all other sub-categories. The sub-categories analyzed

pertain to external factors that inﬂuence driver behav-

ior and/or attention.

4.2.2 Discretization

Next, to achieve the objectives of this article, it will

be necessary to transform the data into a formal con-

text, using the database as input for concept extraction

algorithms. For this purpose, the data should corre-

spond to binary attributes. Not all attributes have this

characteristic, as some are of a numeric and/or cate-

gorical type, within the range [1, 5], where the varia-

tion pertains to how much a driver agrees with a state-

ment (Table 2).

For the transformation of this data into a formal

context, an algorithm was used that considers value

ranges and separates them into two groups (binary),

indicating whether the data agrees or disagrees with

the question. A slight change in the algorithm is when

it is necessary to invert the values 0 and 1. In some

Table 2: Numeric Value and Meaning.

Valor Description

1 Strongly Disagree

2 Disagree

3 Neutral

4 Agree

5 Strongly Agree

cases, an attribute may yield an interesting rule if it is

marked as 1, not necessarily because it belongs to the

group containing values corresponding to 0.

After binarizing the values using the algorithm,

a formal context is obtained that will allow the ex-

traction of rules and implications from the database

(Table 3). This formal context was varied according

to the attributes of interest, separating them into dif-

ferent analyses. This was done due to the limitation

on the number of attributes that FCA algorithms can

handle. In each analysis, the maximum number of at-

tributes was 13, considering the Cartesian product of

attributes from different general categories with sub-

categories.

For the generation of rules, the Lattice Miner 2.0

tool will be used, a data mining prototype developed

under the supervision of Professor Rokia Missaoui

by the laboratory at the University of Quebec. This

is a publicly available Java-based platform where the

main functions include all low-level operations that

allow the manipulation of input data, structures, and

rule association. The platform enables the generation

of groups, called formal concepts, including logical

implications, thus showing binary relationships be-

Application of Formal Concept Analysis to Characterize Driving Behaviors and Socio-Cultural Factors Related to Driving

Table 3: Part of formal context.

less than 5 years of driver’s license

exceeds speed limit

drives under the inﬂuence of friends

drinks while driving

less than 40 years of age

has a college degree

drives carefully when with family

trusts driving skills

X X X X X

X X X

X X

tween collections of objects and their sets of attributes

or properties.

5 RESULTS AND DISCUSSION

When AFC was used, support values were set above

40% and conﬁdence values above 60% in the Chi-

nese drivers’ scenario. The ﬁrst analysis was per-

formed using sociocultural factors (age, educational

factor, and income) along with driving factors (time

spent driving, conﬁdence in driving skills, driving fre-

quency, etc.), and aggressive driving behavior factors

(exceeding speed limits, maintaining safe distances,

etc.), generating two hundred and forty (240) rules.

It is possible to collect these results in XML for-

mat and analyze the rules in the form of A → B. As-

suming that B is a consequence, generating support

(sup) and conﬁdence (conf) (Table 4) generates im-

plication rules pointing to attributes.

When we bring the ”Mindsponge Theory” into

analysis, which explores how external factors inﬂu-

ence driver attitudes, we aim to examine these factors

through two analyses: the inﬂuence of friends and/or

close individuals on driving behavior and the inﬂu-

ence of family on driving conduct. When considering

the former, attributes like ”the inﬂuence of alcohol

and drugs on driving” and ”whether or not there are

people in the car” are taken into account. Therefore,

the database was analyzed for external factors using

AFC, with a minimum support of 55% and a mini-

mum conﬁdence of 88%, resulting in the extraction

of four hundred and sixty rules. The key rules can be

observed in Table 5.

These rules show an interesting characteristic of

the database, leading to the understanding that around

60% of the drivers present drive better and more cau-

tiously when accompanied by of friends and/or close

individuals, with a conﬁdence level of approximately

88%.

Considering the Chinese socio-cultural context,

the three rules found primarily emphasize the impor-

tance of external factors in decision-making and in-

dicate that a signiﬁcant portion of drivers who claim

to have aggressive behavior do not have this kind of

inﬂuence while driving.

On the other hand, an analysis was conducted

considering the category that references family inﬂu-

ence during driving. This analysis considered socio-

cultural aspects along with factors related to family

members in the car, along with the driver. One hun-

dred and ﬁve (105) rules were extracted, with the

main ones documented in Table 6.

In this way, the main extracted rules show that

around 59% of drivers tend not to have aggressive be-

haviors inﬂuenced by the family, with a conﬁdence

level of approximately 87%. With these rules, it can

be afﬁrmed that drivers who have family inﬂuence are

generally less prone to exhibiting aggressive behavior

on the road.

6 CONCLUSIONS

This paper employs FCA, a novel method for exam-

ining driving behaviors from a large database, with

features that allow the application of this method

grounded in the theoretical framework of the ”Mind-

sponge Theory”. This approach is not commonly uti-

lized in the ﬁeld of intelligent transportation and road

safety research.

The association of rules demonstrates characteris-

tics of aggressive behaviors associated with external

factors such as family and friends. It can be inferred

that this is an important factor in a driver’s decision-

making and may be a crucial factor in whether or

not accidents occur. These rules can be generalized

to the Chinese socio-cultural context, and efforts can

be made to understand the primary reasons why this

problem occurs.

On the other hand, it becomes difﬁcult to accu-

rately characterize more speciﬁc individual aspects

due to subjectivity and bias involved in self-reported

survey data.

We had a geographical scope limitation as the

dataset only includes data from Chinese drivers which

may limit the generalizability of our research ﬁnd-

HEALTHINF 2024 - 17th International Conference on Health Informatics

Table 4: Rules extracted considering only aggressive behaviors would be like.

N. Rules Sup Conf

IF A driver has less than 5 years of driving experience

and tends to drive cautiously

THEN they will not exceed the speed limit of the road.

49% 88%

IF A driver has less than 5 years of driving experience

and is under 40 years old

THEN they will yield the right of way to other drivers

40% 90%

IF The driver trusts their skills behind the wheel

and has more than 5 years of driving experience

THEN they will use their turn signal when changing lanes.

44% 80%

Table 5: Rules extracted considering the inﬂuence of friends and/or close individuals.

N. Rules Sup Conf

IF the driver has friends who

inﬂuence not drinking

THEN they will not exhibit aggressive behavior

60% 90%

IF the driver, even in a hurry, has friends

who encourage safe driving

THEN they will yield in trafﬁc

61% 87%

IF the driver has friends who recommend

slowing down at the yellow signal

THEN they will slow down at the yellow signal

60% 87%

Table 6: Rules extracted considering family inﬂuence.

N. Rules Sup Conf

IF mhe driver has their family in the car

THEN they will not exhibit aggressive behavior

60% 86%

IF the driver is criticized by the family for irresponsible driving

THEN they will not exceed the speed limit of the road

58% 88%

IF the driver’s trafﬁc behavior is monitored by the family

THEN they will not run yellow signals

59% 88%

ings. So, for future work, it will be necessary to

explore different scenarios and use different tools to

expand this analysis from the Chinese socio-cultural

context to a more general context, seeking impli-

cations that can lead to better understanding of the

causes of trafﬁc accidents.

ACKNOWLEDGEMENTS

The present work was carried out with the sup-

port of Fundac¸

ao de Amparo

a Pesquisa do Es-

tado de Minas Gerais (FAPEMIG) under grant num-

ber APQ-01929-22. The authors thank CNPq, the

Pontif

ıcia Universidade Cat

olica de Minas Gerais –

PUC-Minas and Coordenac¸

ao de Aperfeic¸oamento

de Pessoal de N

ıvel Superior – CAPES – (Grant

PROAP 88887.842889/2023-00 – PUC/MG, Grant

PDPG 88887.708960/2022-00 – PUC/MG - IN-

FORM

ATICA and Finance Code 001).

REFERENCES

Biedermann, K. (1997). How triadic diagrams represent

conceptual structures. In Lukose, D., Delugach, H.,

Keeler, M., Searle, L., and Sowa, J., editors, Concep-

tual Structures: Fulﬁlling Peirce’s Dream, pages 304–

Application of Formal Concept Analysis to Characterize Driving Behaviors and Socio-Cultural Factors Related to Driving

317, Berlin, Heidelberg. Springer Berlin Heidelberg.

Blevente Lorand Kis, C. S. and Troanca, D. (2017). Fca

tools bundle – a tool that enables dyadic and tri-

adic conceptual navigation. In Proceedings of the In-

ternational Conference on Formal Concept Analysis,

ICFCA ’17, pages 214–219.

Carullo, G., Castiglione, A., De Santis, A., and et al. (2015).

A triadic closure and homophily-based recommenda-

tion system for online social networks. World Wide

Web, 18(6):1579–1601.

Deivid Santos, Cristiane Nobre, L. Z. M. S. (2022). Ap-

plication of formal concept analysis and data mining

to characterize infant mortality in two regions of the

state of minas gerais.

Felde, M. and Stumme, G. (2023). Triadic exploration and

exploration with multiple experts. Knowledge & Data

Engineering Group, University of Kassel, Germany.

Ganter, B. and Obiedkov, S. (2004). Implications in tri-

adic formal contexts. In Wolff, K. E., Pfeiffer, H. D.,

and Delugach, H. S., editors, Conceptual Structures

at Work, pages 186–195, Berlin, Heidelberg. Springer

Berlin Heidelberg.

Houston, J. M., Harris, P. B., and Norman, M. (2003). The

aggressive driving behavior scale: Developing a self-

report measure of unsafe driving practices. North

American Journal of Psychology, 5:193–202.

Hu, B., Dasmahapatra, S., Dupplaw, D., Lewis, P., and

Shadbolt, N. (2004). Managing patient record in-

stances using dl-enabled formal concept analysis. In

Motta, E., Shadbolt, N. R., Stutt, A., and Gibbins,

N., editors, Engineering Knowledge in the Age of the

Semantic Web, pages 172–186, Berlin, Heidelberg.

Springer Berlin Heidelberg.

Jin, R., Wang, X., Nguyen, M.-H., La, V.-P., Le, T.-T., and

Vuong, Q.-H. (2023). A dataset of chinese drivers’

driving behaviors and socio-cultural factors related to

driving. Data in Brief, 49:109337.

Kent, R. E. and Neuss, C. (1997). Conceptual analysis of

hypertext, pages 70–89. Springer Berlin Heidelberg,

Berlin, Heidelberg.

Lucas Ferreira, Cristiane Nobre1, L. Z. M. S. (2021). Study

of the evolution of antiemetic treatment through the

application of triadic formal concept analysis.

Paulo Lana, Cristiane Nobre, L. Z. M. S. (2022). For-

mal concept analysis applied to a longitudinal study

of covid-19.

Wei, L., Qian, T., Wan, Q., and et al. (2018). A research

summary about triadic concept analysis. Interna-

tional Journal of Machine Learning and Cybernetics,

9(4):699–712.

World Health Organization (2023). Road safety.

Yang, J., Du, F., Qu, W., Gong, Z., and Sun, X. (2013).

Effects of personality on risky driving behavior and

accident involvement for chinese drivers. Trafﬁc Inj

Prev, 14(6):565–571.

HEALTHINF 2024 - 17th International Conference on Health Informatics