Identification of Learning Characteristics Pattern of Engineering

Students using Clustering Techniques

Aisyah Larasati

, Apif Miftahul Hajji

and Anik Nur Handayani

Department of Industrial Engineering, Faculty of Engineering, Universitas Negeri Malang, Malang, Indonesia

Department of Civil Engineering, Faculty of Engineering, Universitas Negeri Malang, Malang, Indonesia

Department of Electrical Engineering, Faculty of Engineering, Universitas Negeri Malang, Malang, Indonesia

Keywords: Learning Characteristics, Engineering Students, Data Mining Clustering Technique.

Abstract: Everyone has their own characteristic way of thinking that make them to have different ways to act. These

characteristics also affect their behaviour in daily life, including their learning characteristics. This study aims

to identify the learning characteristics pattern of engineering students using data mining clustering technique.

This study uses questionnaire to collect data. The total number of students fill out the questionnaire are 2,934.

After data preparation steps, only 1,914 responses (65.23% usable rate) are complete and can be used for

further analysis. To identify the learning characteristics pattern, this study uses data mining clustering

technique. The clustering techniques used in this study are K-means cluster, Kohonen cluster analysis, and

two step cluster analysis. The results show that all three cluster techniques used in this study identify the

frequency of a respondent does an independent study by solving practice exercise after learning a new material

in the class, the frequency of a respondent studies the material he learnt after attending a class and the

frequency of a respondent discusses the learning material are the top three important variables to differentiate

each cluster.

1 INTRODUCTION

Everyone has their own characteristic way of

thinking. These characteristics make them to have

different ways to solve a problem and a special

process to identify an issue. Of course, this also affect

their behaviour in daily life, including their learning

characteristic. How they react to many issues in daily

or what they choose to make some decisions on

learning process. It creates human to become a

complex creature. The human behaviour itself is a

human process to act and interact with each other. It

always becomes a complex process that hard to

understand, considering human behaviour depends on

many life factors. Human behaviour can determine

people to take some decision and create a habit on

daily basis. According to Icek Ajzen theory, human

behaviour guided by three consideration: beliefs

about the result of their behaviour and the evaluation

of their result (behavioural beliefs), beliefs about

other people expectations and motivate to do it

(normative beliefs), and beliefs about factors that can

inhibit or facilitate the accepted behaviour and the

impact of it (control beliefs) (Ajzen, 1985). What

people experience in daily life will create their

behaviour pattern. People will show their behaviour

with their actions or how they interact with each

other.

The world of education is also changing due to the

global internet phenomenon. Internet gives people

much accessible media for learning source. A large

number of accessible media changes college student

behaviour to learn. Basically, students have their own

reason to choose how they learn, what their learning

styles, which media will be used, etc. Students media

usage behaviour is strongly influenced by three

factors: sociability, utility and reciprocity (Zawacki-

Richter et al., 2015). Sociability can be reviewed by

their interaction with each other which lead to

selected media. The utility is processing to get the

best result with by using accessible media with

maximal effort. The last, reciprocity is how they use

accessible media to improve their cognitive skill by

reading, being critical, and understanding. Of course,

not all of them choose internet for media to learn. A

few college students still feel better to understand

what they learn with a book or any non-internet

media.

274

Larasati, A., Hajji, A. and Handayani, A.

Identiﬁcation of Learning Characteristics Pattern of Engineering Students using Clustering Techniques.

DOI: 10.5220/0008411002740278

In Proceedings of the 2nd International Conference on Learning Innovation (ICLI 2018), pages 274-278

ISBN: 978-989-758-391-9

The difference in learning characteristics is

influenced by individual characteristics as well. In

addition to behaviour, personal characteristics play a

major role in the learning characteristics. Personal

characteristics show special behaviour in each

individual. Various studies have revealed the

importance of understanding student characteristics

on the effectiveness of the learning process

(Kauffman, 2015; VanSickle et al., 2015; Apple,

Duncan and Ellis, 2016). However, there is still not

much literature that explore student characteristics by

applying data mining technique. According to (Sin

and Muthu, 2015), data mining techniques can be

used to improve academic quality, including

predicting student performance in learning, data

visualization, detecting student failures in learning

and even investigating student behaviour in learning.

Therefore, data mining techniques are also potential

to be used to segment student characteristics in order

to understand their learning behaviour. This study

aims to identify the learning characteristics pattern of

engineering students using data mining clustering

technique. The cluster resulted from this research can

be used to figure out the existing differences among

cluster and provide faculty members with some

insight of their student characteristics.

2 METHOD

This study uses an online questionnaire to collect

data. The online questionnaire is administered via

Universitas Negeri Malang Academic Information

System (SIAKAD) in April - May 2018. The target

respondents are all registered students in Faculty

Engineering at Universitas Negeri Malang, which are

approximately 5,300 students. Among of those

registered students, only 2,934 students fill out the

questionnaire (55.34% participation rates). The data

mining clustering model is built following the

SEMMA procedure, which are Sample, Explore,

Modify, Model, and Assess.

The first step, sample is conducted by determining

the target object of the study, which are all registered

students in Universitas Negeri Malang. The explore

step aims to understand the nature of collected data,

which is performed by plotting the collected data. The

modify step includes data preparation and data

transformation when needed. Data preparation steps

include data cleaning and data imputation. Data

cleaning aims to delete uncompleted responses and

outlier responses. Data imputation is performed to

impute missing responses with the mode or mean

responses. After data preparation steps, only 1,914

responses (65.23% usable rate) are complete and can

be used for further analysis. To identify the learning

characteristics pattern, this study uses three clustering

data mining technique. The clustering models built in

this study are K-means cluster, Kohonen cluster

analysis, and Two step cluster analysis. The last step

is to determine how to assess the model performance

(accuracy). Regarding the model accuracy, this study

use Silhouette index as suggested by (Pereda and

Estrada, 2018). Silhouette measures distance of an

element to its own cluster (cohesion) and compares it

to other clusters (separation). The higher value

indicates that the element is well matched to its own

cluster and poorly matched to other defined clusters.

The questionnaire that is used to collect data

contains 4 questions about respondent’s profile

(gender, level of study period, study program, and

GPA). In addition, the questionnaire contains 18

closed-ended question that ask about the learning

characteristic of respondent. Briefly, the item list of

the questions in the questionnaire is shown in Table

Table 1: Item list in the questionnaire.

Variable

Indicators

Number of

Items

Learner

Characteristics

Personal Profile

4 open-

ended

questions

Learning preparation

1, 2

Initial understanding

3, 4, 5, 6

Characteristics of

discussion activity

7, 8, 9, 10,

Understanding during

learning process

12, 13, 15

Characteristics of

independent study

14, 16, 17

Learning styles

Table 2: Descriptive statistics of the respondent’s profile.

Department

Count

(respondent)

Percentage (%)

Civil

Engineering

423

22.10%

Electrical

Engineering

671

35.06%

Mechanical

Engineering

500

26.12%

Industrial

Technology

320

16.72%

Identiﬁcation of Learning Characteristics Pattern of Engineering Students using Clustering Techniques

275

The descriptive statistic of the personal

respondents profile is shown in Table 2. In addition

to the department of the respondents, the profile also

show that the respondents are students in the first

years up to the seventh years of study in the faculty of

engineering who has GPA from 2.00 – 4.00.

3 RESULT AND DISCUSSION

The results of the classification using data mining

techniques as shown in Table 3 indicate that K-means

technique results in the highest number of cluster,

meanwhile Two Steps techniques results in the lowest

number of cluster. However, the Kohonen technique

results in the highest range of the size of cluster.

Based on the range of the cluster size, it seems that

Kohonen technique works best to classify students

learning characteristics. This results implies that it is

better to classify the learning characteristics into five

clusters: cluster 1 (20.34%), cluster 2 (30.46%),

cluster 3 (6%), cluster 4 (14.54%), and cluster 5

(28.66%). The largest cluster, cluster 2, is dominated

by students who are moderate frequently do an

independent study. On the other hand, cluster 4, is

dominated by student who are always do an

independent study. While in other cluster, the

students are rarely do an independent study.

The model accuracy is measured based on the

silhouette cohesion and separation index. The

silhouette indices of the three clustering models in

this study imply that the best clustering method is the

two steps model since this model results in the highest

Silhouette index. On the other hand, K-means cluster

has the lowest Silhouette index. This indices indicate

that the Two steps model has the best cohesion and

separation ability compared to K-means and Kohonen

models.

Table 3: Comparison of the classification results using

different techniques

Model

Numb

er of

Cluste

Smalles

Clusters

(%)

Largest

Cluster

Silhouett

e Index

K-means

0.21

Kohonen

0.63

Two steps

0.82

Detail results shown at Figure 1, Figure 2, and

Figure 3 indicate that all three cluster techniques used

in this study identify var 14 (how frequent a

respondent does an independent study by solving

practice exercise after learning a new material in the

class), var 15 (how frequent a respondent studies the

material he learnt after attending a class) and var 11

(how frequent a respondent discusses the learning

material) are the top three important variables to

differentiate each cluster.

An independent study, either by doing some

exercise or studying learning material, is a form of an

active learning activity. An independent study

improve student performance in undergraduate

science, technology, engineering, and mathematics

(STEM) course (Freeman et al., 2014). Students’

intention to do independent study may vary across all

department in Faculty of Engineering. Thus, solving

practise exercise and repeating studying learning

material become important variable to cluster

students learning characteristics.

Discussion in learning process is an activity that

requires a student to express his/her thought to other

and gain feedback on it. Discussion activity include

two main process, an external interaction and an

internal process (Illeris, 2009). The external

interaction means a student have to interact to his/her

teacher/peer, or surrounding environment. The

internal process related to his/her psychological

ability to elaborate and acquire information into

his/her mind. Discussion intention might be

significantly differently among the respondents, thus

it become one of the top three important variables to

cluster them. This result is also in line with other

study that find relationship between frequency of

discussion with tutor and learning outcomes.

Learning characteristics, which include self-efficacy

and inside knowledge affects the structure of the

discussion (Mitchell et al., 2013).

Figure 1: Cluster size and predictor importance based on k-

means technique.

ICLI 2018 - 2nd International Conference on Learning Innovation

276

Figure 2: Cluster size and predictor importance based on

kohonen technique.

In addition, the three cluster techniques used in

this study do not include var 2 (a respondent motive

to do learning preparation before attending a class),

var 3 (a respondent learning style while developing

early understanding of the learning material), var 5

(the reason behind a respondent behaviour to develop

early understanding of the learning material) and var

6 (a respondent learning style to prepare her/himself

before attending a class) as important variables to

cluster the respondents. It means that there is no

significant difference among respondents based on

these four variables. This results implies that the

students’ motive, learning styles, intention to develop

early understanding is vary across all department in

Faculty of Engineering. There is no specific pattern

that may be used to classify the respondent.

Figure 1 shows the least important factor to cluster

the respondent using K-Means technique is gender.

This result implies that female and male are only

slightly different in their learning characteristics

according to K-Means cluster. This result is in

accordance with (Subramanian, 2018) findings that

indicate no differences between male and female

students in their learning styles. Figure 2 indicates

that the least factor to cluster the respondent using

Kohonen technique is var 13 (the activity that a

respondent do when he/she does not understand a

learning material). This finding implies that there is

only slightly different on the student activity that are

performed by the students when they don’t

understand the learning material. The individual

responses show that most students (859 students) tend

to ask their friend instead of doing the other available

options. In addition, the Two Step technique identify

var 17 (a place where a respondent usually does an

independent study) as the least important variable.

This result implies that students’ preferences on

choosing the place to study is only slightly different

among cluster. It is supported by the individual

responses that reveal the study area provided in the

campus is the most favourite place to study.

Figure 3: Cluster size and predictor importance based on

two steps technique.

4 CONCLUSIONS

Based on Silhouette index, the best model to cluster

student learning characteristics is the Two Step

Model. The top three students learning characteristics

that are important to differentiate one cluster to the

other cluster are: 1) the frequency of a student does

an independent study by solving practice exercise

after learning a new material in the class; 2) the

frequency of a student studies the material he learnt

after attending a class; and 3) the frequency of a

student discusses the learning material. On the other

hand, the other learning characteristics that are only

slightly different between one cluster to the other

clusters are: 1) a student motive to do learning

preparation before attending a class; 2) a student

learning style while developing early understanding

of the learning material, the reason behind a student

behaviour to develop early understanding of the

learning material; and 3) a student learning style to

prepare her/himself before attending a class.

ACKNOWLEDGEMENTS

Acknowledgments to Universitas Negeri Malang

(UM) which has funded this research through Islamic

Development Bank (IsDB) - UM Research Grant.

REFERENCES

Ajzen, I. (1985) ‘From Intentions to Actions: A Theory of

Planned Behavior’, in Kuhl, J. and Beckmann, J. (eds)

Action Control: From Cognition to Behavior. Berlin,

Heidelberg: Springer, pp. 11–39.

Identiﬁcation of Learning Characteristics Pattern of Engineering Students using Clustering Techniques

277

Apple, D., Duncan, W. and Ellis, W. (2016) ‘Key Learner

Characteristics for Academic Success’, International

Journal of Process Education, 8(2).

Freeman, S. et al. (2014) ‘Active learning increases student

performance in science, engineering, and mathematics’,

Proceedings of the National Academy of Sciences,

111(23), pp. 8410–8415.

Illeris, K. (2009) ‘A comprehensive understanding of

human learning’, in Illeris, K. (ed.) Contemporary

Theories of Learning: Learning Theorists … In Their

Own Words. New York: Routledge, pp. 1–14.

Kauffman, H. (2015) ‘A review of predictive factors of

student success in and satisfaction with online

learning’, Research in Learning Technology, 23.

Mitchell, C. M. et al. (2013) ‘Learner characteristics and

dialogue: recognising effective and student-adaptive

tutorial strategies’, International Journal of Learning

Technology, 8(4), pp. 382–403.

Pereda, M. and Estrada, E. (2018) ‘Machine Learning

Analysis of Complex Networks in Hyperspherical

Space’, eprint arXiv:1804.05960.

Sin, K. and Muthu, L. (2015) ‘Application of big data in

education data mining and learning analytics--a

literature review’, ICTACT Journal on soft computing,

5(4).

Subramanian, S. (2018) ‘Learning styles and achievement

in english among higher secondary school students–

gender wise analysis’, Indian Journal Of Applied

Research, 8(8).

VanSickle, J. L. et al. (2015) ‘An Investigation of Student

Performance, Student Satisfaction, and Learner

Characteristics in Online Versus Face-to-Face Classes-

-RESEARCH’, Kentucky Journal of Excellence in

College Teaching and Learning, 13(1), p. 3.

Zawacki-Richter, O. et al. (2015) ‘Student media usage

patterns and non-traditional learning in higher

education’, The International Review of Research in

Open and Distributed Learning, 16(2), pp. 136–170.

ICLI 2018 - 2nd International Conference on Learning Innovation

278