Association Rule Mining with Apriori Algorithm for Pediatric Foot

Disorders

1Department of Healthcare Engineering, Graduate school

Chonbuk National University

664-14, Deokjin-dong 1, Deokjin-ku, Jeonju, Jeolabuk-do, 561-756

REPUBLIC OF KOREA

2School of Electronics and Computer Engineering

Chonnam National University

77 Yongbong-ro, Buk-gu, Gwangju, 500-757

REPUBLIC OF KOREA

3Division of Biomedical Engineering

Chonbuk National University

664-14, Deokjin-dong 1, Deokjin-ku, Jeonju, Jeolabuk-do, 561-756

REPUBLIC OF KOREA

4Research Center of Healthcare & Welfare Instrument for the Aged

Chonbuk National University

664-14, Deokjin-dong 1, Deokjin-ku, Jeonju, Jeolabuk-do, 561-756

REPUBLIC OF KOREA

Abstract: - An association rule mining has potential to discover important information of disorder in data. The

purpose of this study was to show analysis process of the clinical data to acquire significant information

effectively between the foot disorder groups and biomechanical parameters related to symptom by the

association rule discovery. The first clinical health records of the total 279 pediatric patients diagnosed with the

complex foot disorder, including pes planus basically, were used for study. The Apriori algorithm was applied

to discover rule of the foot disorder groups. As the results, we were able to discover 8 rules of the complex foot

disorder and confirm major information. In next study, another data mining methodology like neural network

will be applied with a careful preprocessing for better analysis of the pediatric foot disorder from now on.

Key-Words: Data mining, Association rule, Apriori, Pediatric foot, Foot disorder

1 Introduction

In medical field, massive data sets like the

electronic health records is created by fast

development of hospital information technology.

The clinical data includes quantitative data (e.g.

laboratory values), qualitative data (e.g. text-based

documents and demographics), and transactional

data (e.g. record of medication delivery) [1]. When

utilization of medical big data, value production of

330 billion dollars is expected every year on the US

medical field. If effective treatment method by

analysis data of diagnostic pattern, prognosis, cost,

etc., direct effect of about 165 billion dollars is

expected [2]. Data mining is the extraction

technology of hidden and valuable information, and

it has been recognized for a large amount of the

clinical data by many study [3]. It is method to deal

with large data and to discover meaningful

knowledge with application of pattern recognition

technology, statistics or mathematic algorithm [4].

An association rule discovery, is first introduced by

Agrawal, Imieliński, and Swami, has potential to

discover important information of disorder in data

[5]-[7]. It is the most valid methodology to explore

attributes pairs which contain useful knowledge in

nonrestrictive patterns. In other words, rules from

this method mean the relation between particular

transaction and other transaction appeared

progressively or simultaneously when specific

transaction occurred. Accordingly, the advantage of

JUNGKYU CHOI1, YONGGWAN WON2, JUNG-JA KIM3,4,*

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2022.21.9

Jungkyu Choi, Yonggwan Won, Jung-Ja Kim

E-ISSN: 2224-2872

Volume 21, 2022

the association rule discovery is to be able to

analyze a correlation of variables in data at the same

time by searching items to set of rules.

According to previous studies, the data mining was

applied for analysis of the clinical data. Kim

presented that age, pathology scale, related disease,

hospitalization period, respiratory failure and

congestive heart failure were came out to be danger

factors on death of pneumonia by using Apriori

modeling algorithm in study for analysis of death

factor on pneumonia patient [8]. Breault, Goodall

and Fos utilized Classification and Regression Trees

(CART) of data mining for analysis of diabetic data

warehouse. They figured out that the most important

variable associated with bad glycemic control was

younger age, not the comorbidity index or whether

patients had related disorders [9]. In addition, Duru

used four decision tree algorithms for analysis

postoperative status in the ovarian endometriosis

patient under different conditions, and reported new

meaningful information about recurrent ovarian

endometriosis [10]. However, these data mining

technology was not tried to utilize yet in terms of

podiatric medicine. So far the earlier studies have

verified significance only through a simple

statistical method.

The lower limbs are the most important organ for

movement action of human and basic activity of

daily living [11]. Most of all, the foot have 26

bones, 33 joints, more than 100 muscles, tendons

and ligaments, even if it is just 5% of the entire

surface in the body. In addition, it works organically

as a highly complex anatomical- and biomechanical

structure in gait with supporting the whole weight

and ground reaction force [12]. These pressure,

weight and ground reaction force by push-off

exercise, causes stress or soft tissue strain [13]. In

this condition, there are close connections between

form of the foot and disorder the lower limbs.

However, complex correlation analysis is necessary

than simple quantitative analysis between the foot

disorder and symptom because the foot disorder

appear with complex symptoms, and this symptoms

are not commonly clear.

Accordingly, the purpose of this study was to show

analysis process of the clinical data to acquire

significant information effectively between the foot

disorder groups and biomechanical parameters

related to symptom by the association rule

discovery.

2 Study Procedure

2.1 Subjects

The first clinical health records of the total 279

pediatric patients diagnosed with the complex foot

disorder, including pes planus basically, were used

for study. The data was collected from the Foot

Clinic of Jeon-ju Pediatrics. 64 patients records with

missing values were excluded, and the complex

disorder groups over 5% of whole data were

selected. Therefore, analysis data was composed of

174 patient records with five groups for the complex

disorder.

2.2 Item sets

A consequent in the study was the foot disorder, and

it was into encoding for more clear classification.

Among them, the consequent in the analysis data

was consisted of five complex disorder groups such

as class A : D1 (Achilles tendinitis), D2 (Pes

planus), class B : D2 (Pes planus), class C : D2 (Pes

planus), D3 (Intoe gait), class D : D2 (Pes planus),

D3 (Intoe gait), D5 (Genu valgum), class E : D2

(Pes planus), D5 (Genu valgum). The foot disorder

commonly appeared complexly like this. An

antecedent was the biomechanical parameters

related to disorder, and it was 26 attributes at first.

To check statue of the foot, the biomechanical

parameters like the Resting Calcaneal Stance

Position (RCSP), the tibia TransMalleolar Angle

(TMA) or the knee Internal Malleolus Distance

(IMD) were measured and recorded into a patient

chart by a podiatrist, as shown in Fig. 1. T hrough

the feature selection node, meaningless antecedents

were excluded. The extracted 13 attributes were

grouped for changing numeric type into nominal

type based on Donatelli’s clinical assessment of the

foot by radiography and Choi, Kim, Won and Kim’s

result of decision tree model [14]-[15]. Finally, 7

antecedents were used for analysis, as follow Table

Fig 1. Measurement of RCSP and patient charts

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2022.21.9

Jungkyu Choi, Yonggwan Won, Jung-Ja Kim

E-ISSN: 2224-2872

Volume 21, 2022

2.3 Data analysis

In the study, combination of the antecedent meant

each of the foot disorder group. Therefore, 7

antecedents were inserted into Apriori algorithm of

association rule mining for the consequent. Data

analysis was performed by IBM SPSS statistics 18

(SPSS Inc., Chicago, IL, USA) and IBM SPSS

Modeler 14.2 (SPSS Inc., Chicago, IL, USA). In the

Apriori algorithm, a assocation rule (R) of the

antecedent (X) cause the consequent item (Y), and

this rule can be expressed as follows :

(R) : (X) ⇒ (Y) (1)

For significance, a rule which satisfied with the

minimum support and the minimum confidence set

up by a user is only extracted. In this study, the

Minimum Support (MS) was 5 %, and the Minimum

Confidence (MC) was 85 %. The overall study

process was shown in Fig. 2.

Table 1. Consequent Item

Class

Disorder

D1, D2

D2, D3

D2, D3, D5

D2, D5

Fig 2. Study Process

3 Result

The first clinical health records of 174 pediatric

patients with the complex foot disorder, including

pes planus basically, were used for this study in the

279 whole data. The Apriori algorithm was applied

to the experimental data to discover rule of the foot

disorder groups.

As a result, total 8 rules which were satisfied

with the minimum support and the minimum

confidence were generated, as follows : Rule 1 :

KneeIMD was abnormal and TibiaTMA was

abnormal ⇒ Class D (MS : 14.451%, MC : 92%),

Rule 2 : KneeIMD was abnormal, TibiaTMA was

abnormal and RCSP was 1 group ⇒ Class D (MS :

12.139%, MC : 90.476%), Rule 3 : KneeIMD was

abnormal, TibiaTMA was abnormal and

TalarDeclination was 2 group ⇒ Class D (MS :

12.717%, MC : 90.909%) Rule 4 : Talocalcaneal

was 3 g roup, CuboidAbduction was 2 g roup and

Intermetatarsal was 2 group ⇒ Class B (MS :

9.249%, MC : 93.75%), Rule 5 : CuboidAbduction

was 1 group, TibiaTMA was abnormal and RCSP

was 1 group ⇒ Class C (MS : 16.763%, MC :

86.207%), Rule 6 : KneeIMD was abnormal,

TibiaTMA was abnormal, RCSP was 1 group and

TalarDeclination was 2 group ⇒ Class D (MS :

10.405%, MC : 88.889%), Rule 7 : Intermetatarsal

was 1 group, TibiaTMA was abnormal, RCSP was 1

group and TalarDeclination was 2 group ⇒ Class C

(MS : 12.717%, MC : 86.364%), Rule 8 :

CuboidAbduction was 1 group, TibiaTMA was

abnormal, RCSP was 1 group and TalarDeclination

was 2 group ⇒ Class C (MS : 15.607%, MC :

88.889%), as shown in Fig. 3. In case of class A and

class E, reliable rules were not discovered.

4 Conclusion

The purpose of this study was to perform analysis

process of the clinical data to acquire significant

information between the foot disorder groups and

biomechanical parameters related to symptom by

the association rule discovery. The first clinical

health records of 174 pediatric patients with the

complex foot disorder were analyzed for this study,

in the 279 whole data. Through preprocessing like

grouping, the experimental data was composed of a

consequent, five classes, and 7 antecedent. The

Apriori algorithm was applied to the experimental

data to discover rule of the foot disorder groups.

As the results, we were able to discover 8 rules of

the complex foot disorder and confirm major

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2022.21.9

Jungkyu Choi, Yonggwan Won, Jung-Ja Kim

E-ISSN: 2224-2872

Volume 21, 2022

Fig 3. Result of Association Rules

information. In case of the class D, pes planus, intoe

gait and genu valgum, there was possibility to occur

with high probability if the knee internal malleolus

distance and the tibia transmalleolar angle were

abnormal, RCSP was below -6° and the talar

declination angle was over 22°. The class C, pes

planus and intoe gait, was likely to appear highly in

condition of that the tibia transmalleolar angle was

abnormal, RCSP was below -6°, the talar

declination angle was over 22° and the

cuboidabduction angle was below -1°, or the

intermetatarsal angle was below -7°. As the result of

class C and D, especially, we were able to know that

key factor which decided genu valgum was the knee

internal malleolus distance. In case o f the class B,

pes planus, there is likely to appear highly if the

talocalcaneal angle was asymmetry between both

feet, the cuboid abduction angle was over 6° and the

intermetatarsal angle was over 9°. Generally, the pes

planus was diagnosed by the resting calcaneal

stance position angle [16]. All of classes had pes

planus basically because the highest value of RCSP

was just -4° in the experimental data. However, the

result of the class B showed a new rule with other

parameters, instead of RCSP. In other words, it

meant that RCSP was a factor for diagnosis of pes

planus-nothing more and nothing less.

A massive amount of medical data is available now

with hospital system, and an intelligent analysis

method is necessary to get meaningful information

effectively in complex data [17]. However, these

efforts were not attempted yet in the podiatric

medicine field, and the pattern analysis of foot

disorder in pediatric patient was particularly

insufficient. Through this study, we were able to get

useful information with the result that we tried to

discover rule of the podiatric foot disorder.

However, rules of two classes were not confirmed.

This result would be judged by reason of that the

foot disorder had commonly complex symptoms and

the symptoms were not clear. This being so, more

detailed data preprocessing like grouping is very

important in the foot clinical data. In next study,

another data mining methodology like neural

network will be applied with a careful preprocessing

for better analysis of the pediatric foot disorder from

now on.

Table 2. Antecedent Item

Variable

Type

Value

Description

RCSP Nominal 1: below -6°, 2: over -5°, 3: asymmetry

Resting calcaneal

stance position angle

Tibia TMA Flag 1: abnoraml, 0: normal

Angle of the tibia

transmalleolar

Knee IMD Flag 1: abnoraml, 0: normal

The knee internal

malleolus distance

Talocalcaneal Nominal 1: below 28°, 2: over 27°, 3: asymmetry

Angle between the talus

and the calcaneus

Cuboidabduction Nominal 0: normal, 1: below -1°, 2: over 6°, 3: asymmetry

Angle of the cuboid

abduction

Intermetatarsal Nominal 0: normal, 1: below -7°, 2: over 9°, 3: asymmetry

Angle of the metatarsus

primus adductus

Talardeclination Nominal 1: below 20°, 2: over 22°, 3: asymmetry

Angle of the talus

declination

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2022.21.9

Jungkyu Choi, Yonggwan Won, Jung-Ja Kim

E-ISSN: 2224-2872

Volume 21, 2022

Acknowledgments

This work was supported by the National

Research Foundation of Korea (NRF) grant funded

by the Korea government (MSIP) (NRF-

2013R1A2A2A04016782)

References:

[1] T. B. Murdoch, and A. S. Detsky, “The

inevitable application of big data to health care.

Jama, Jama, vol. 309, no. 13, 2013, pp. 1351-

1352.

[2] J. Manyika, M. Chui, B. Brown, J. Bughin, R.

Dobbs, C. Roxburgh, and A. H. Byers. Big

data: The next frontier for innovation,

competition, and productivity, Mckinsey

Global Institute, June 2011.

[3] A. R. Islam, and T. S. Chung, An Improved

Frequent Pattern Tree Based Association Rule

Mining Technique. Information Science and

Applications (ICISA), 2011 International

Conference on. IEEE, 2011, pp. 1-8.

[4] T. L. Lincoln, and P. W. Suen, Common

rotational variations in children, Journal of the

American Academy of Orthopedic Surgeons,

vol. 11, no. 5, 2003, pp. 312-320.

[5] R. Agrawal, T. Imieliński, and A. Swami,

Mining association rules between sets of items

in large databases, ACM SIGMOD Record, vol.

22, no. 2, 1993, pp. 207-216.

[6] T. Imamura, S. Matsumoto, Y. Kanaqawa, B.

Tajima, S. Matsuya, M. Furue, and H. Oyama,

A technique for identifying three diagnostic

findings using association analysis, Medical &

Biological Engineering & Computing, vol. 45,

no. 1, 2006, pp. 51-59.

[7] W. T. Hwang, and D. S. Kim, Improved

association rule mining by modified trimming,

Journal of the Institute of Electronics

Engineers of Korea, vol. 45, no. 3, 2008, pp.

15-21.

[8] Y. M. Kim, A study on analysis of factors on in-

hospital mortality for community-acquired

pneumonia, Journal of the Korean Data &

Information Science Society, vol. 22, n o. 3,

2011, pp. 389-400.

[9] J. L. Breault, C. R. Goodall, and P. J. Fos, Data

mining a d iabetic data warehouse, Artificial

Intelligence in Medicine, vol. 26, no. 1, 2002,

pp. 37-54.

[10] N. Duru, An application of apriori algorithm on

a diabetic database, Knowledge-Based

Intelligent Information and Engineering

Systems, Springer Berlin Heidelberg, 2005, p.

398-404.

[11] Y. J. Ko, and H. W. Kim. Diagnosis and

Conservative Treatment of Common Foot

Diseases, Journal of the Korean Medical

Association, vol 47, 2004, pp. 247-257.

[12] H. C. Kim, Management of Foot and Ankle

Disorders, Journal of the Korean Medical

Association, vol. 48, 2005, pp. 663-671.

[13] D. J. Lott, M. K. Hastings, P. K. Commean, K.

E. Smith, and M. J. Muller, Effect of footwear

and orthotic devices on stress reduction and

soft tissue strain of the neuropathic foot,

Clinical Biomechanics, vol. 22, 2007, pp. 352-

359.

[14] R. A. Donatelli, The Biomechanics of the foot

and ankle, The F. A. Davis Company,

Philadelphia, PA, 1995.

[15] J. Choi, H. I. Kim, Y. Won, and J. J. Kim,

Pattern Analysis of Pediatric Foot Disorders

Using Decision Tree, Recent Researches in

Applied Computer Science, 2015, pp. 9-14.

[16] W. H. Lee, and S. W. Lee, A study of the

Relationship between Normal Adults Resting

Calcaneal Stance Position and Postural Sway, J

Korean Soc Phys Ther, vol. 11, 2004, pp. 5-17.

[17] W. Raghupathi, Data mining in health care,

Healthcare Informatics: Improving Efficiency

and Productivity, 2010, pp. 211-223.

Creative Commons Attribution License 4.0

(Attribution 4.0 International, CC BY 4.0)

This article is published under the terms of the Creative

Commons Attribution License 4.0

https://creativecommons.org/licenses/by/4.0/deed.en_US

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2022.21.9

Jungkyu Choi, Yonggwan Won, Jung-Ja Kim

E-ISSN: 2224-2872

Volume 21, 2022