You are on page 1of 5

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/324177399

Attribute selection on student performance dataset using maximum


dependency attribute

Conference Paper · October 2017


DOI: 10.1109/ICEEIE.2017.8328784

CITATIONS READS

11 232

5 authors, including:

Rd. Rohmat Saedudin Edi Sutoyo


Telkom University Telkom University
61 PUBLICATIONS   192 CITATIONS    52 PUBLICATIONS   136 CITATIONS   

SEE PROFILE SEE PROFILE

Shahreen Kasim Iwan Tri Riyadi Yanto


Universiti Tun Hussein Onn Malaysia Ahmad Dahlan University
170 PUBLICATIONS   600 CITATIONS    64 PUBLICATIONS   316 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Comparative Study of Different Window Sizes Setting in Median Filter for Off-angle Iris Recognition View project

rough set View project

All content following this page was uploaded by Edi Sutoyo on 16 July 2018.

The user has requested enhancement of the downloaded file.


2017 5th International Conference on Electrical, Electronics and Information Engineering (ICEEIE)

Attribute Selection on Student Performance Dataset


Using Maximum Dependency Attribute

Rd Rohmat Saedudin, Edi Sutoyo Shahreen Kasim, Hairulnizan Iwan Tri Riyadi Yanto
School of Industrial and System Mahdin Department of Information Systems
Engineering Faculty of Computer Science and Universitas Ahmad Dahlan
Telkom University Information Technology, Universiti Yogyakarta, Indonesia
Bandung, West Java, Indonesia Tun Hussein Onn Malaysia yanto.itr@is.uad.ac.id
rdrohmat@telkomuniversity.ac.id, BatuPahat, Johor, Malaysia
edisutoyo@telkomuniversity.ac.id shahreen@uthm.edu.my,
hairuln@uthm.edu.my

Abstract— As a higher education institution, knowing which constantly changing data is student data. Student data need to be
GPA of the semester has the most determinant to affecting the extracted to obtain an important information for the purposes of
academic performance of students is important yet challenging. determining the ratio of lecturer, accreditation, graduation rate,
Therefore, this paper deliberates the usage of rough set theory student performance and others.
based Maximum Dependency Attributes (MDA). The dataset is
taken from the Directorate of Information Systems (SISFO), In educational system, academic performance of the students
Telkom University. The result showed that the most determinant mostly can be evaluated using the grade point average (GPA)
attribute is 2th GPA, followed with 3rd GPA, Entrance score of the students in every semester. As a higher education
Examination, 1st GPA, Academic Aptitude Test, and 4th GPA, institution, knowing which GPA of the semester has the most
respectively. By early knowing the most determinant attribute in determinant to affecting the academic performance of students
which score of the m, a well-planned strategic program can be set is important yet challenging. By early knowing the most
during the institution academic study period. determinant attribute in which GPA of the semester, a well-
planned strategic program can be established during the
Keywords— Rough set theory; Maximum Dependency academic period.
Attributes; Clustering; Student Dataset.
Thus, this article deliberate the usage of rough set theory to
I. INTRODUCTION cluster the attribute selection for student performance dataset is
University student’s performance should be greatly presented. This method was developed based on a coarse set
concerned since it may affected by several factors. For example, approach with maximum dependency attributes invented by
detecting why so many students fail to pass on time is a major [13]. The foremost our work contribution is related to the most
social issue for higher education institutions. By implementing decisive attribute selection by generating the relevant attribute
student performance evaluation, a well-organized platform can rank based on the dataset-maximum attribute dependency.
be delivered for the period of the institution study period [1], Choosing and knowing the most prominent determinant
such as arrange an intensive guidance to improve student’s attributes of the student's GPA in advance can contribute to
academic performance. A potential effective technique to provide recommendation on intervention design, to undertake
identify student failures is using data mining approaches [2]. maintenance, and to increase student academic performance.
In general, data mining in general may execute the process The rest of paper organization are detailed as following
of finding, analyze new information in the existed data then chapters. The second section conveys the basic knowledge of
summarize it as valuable information. Several outstanding information system, rough set theory, Maximum dependency
studies on data mining are in area of clustering, association rules, Attributes (MDA). Section 3 clarifies the detail of selected
classification, and conflict analysis [3–11]. The field of data dataset characteristics. Section 4 deliberates the results of the
mining application in the education area is educational data experiment and the evaluation of the results. Finally, the
mining (EDM). Application of EDM are extensive research conclusion paper is presented in section 5.
areas containing e-learning, intelligent guidance systems, text
II. THEORETICAL BACKGROUND
mining, social network mining, and others. [12]. The EDM
process discovers a useful information coming from the A. Information System
educational systems data that has the potential to have an The information system (IS) concept provides an easy-to-use
impactful thing on educational research. Since in facts, various tool for representation of the object attribute values. The IS has
data in universities continue growing dynamically in line with
academic and non-academic activities in it. One of the a quadruple (four tuples) S = (U , A, V , f ) , where

978-1-5386-0355-0/17/$31.00 ©2017 IEEE

176
2017 5th International Conference on Electrical, Electronics and Information Engineering (ICEEIE)

{ }
U = u1 , u 2 ," , u U is a non-void finite set of objects, D in a degree of k is said to be totally depends on C if k=1.
A = {a , a , " , a }
1 2 A
is a non-void set of attributes, Or else, D is partially depends on C, while PosC is in positive
region.
V = * a∈ A V a , V a is the value set of a, f : U × A → V is an

information function, such that f (u , a ) ∈ Va , for each


Pos C = ¦ X ∈U / D
C (X )

(u , a ) ∈ U × A , known as information function. From the above


definition of IS, the following section may briefly explain the Proposition 1 (See [13]). Let S = (U , A,V , f ) be an IS
rough set theory. and let D and C be any subsets of A. If D entirely depends on C,
then α D ( X ) ≤ α C ( X ) for every X ⊆ U .
B. Rough Set Theory
Professor Zdzislaw Pawlak is the initiator of Rough set
theory [14]. Rough set theory is necessary for representing a Proof. Let D and C be any subsets of A in IS
universe’s subset as equivalence class of the divider of the S = (U , A,V , f ) . The hypothesis stated that
universe. The relationship indiscernibility is the early argument
IND (C ) = IND (D ) . Moreover, the clustering U/C is finer than
for forming partitions. Two elements x, y ∈U in
that U/D. Thus, it is clear that any correspondence class induced
S = (U , A,V , f ) is said to be B-indiscernible if and only if by IND (D ) is a union of some equivalence class induced by
f ( x, a ) = f ( y , a ) , for every a ∈ B . An indiscernible relation IND (C ) . Therefore, for every single x ∈ X ⊆ U , we have
induced by the set of attribute B, denoted by IND (B ) , is a [x ] ⊆ [x ]D . Later, for every X ⊆ U , we have
C
correspondence relation, can encourage a distinctive partition.
The partition of U induced by IND (B ) in S = (U , A, V , f ) D( X ) ⊆ C ( X ) ⊂ X ⊂ C( X ) ⊆ D( X ) .
denoted by U / B and the equivalence class in the partition
U / B contains x ∈ U and denotes by [x ]B . Here, B is defined
Consequently,
as any subset of A in S and X is defined as any subset of U, the
B-lower approximation of X, denoted by B ( X ) and B-upper D( X ) C(X )
α D (X ) = ≤ = α C (X )
approximation of X, denoted by B( X ) correspondingly, D( X ) C(X )
{
B( X ) = x ∈ U [x] B
}
⊆ X and Fig. 1 obviously depicts the method pseudo-code, comprises
B( X ) = {x ∈U [x] } of four core steps. The initial step is used to compute the
B
 X ≠φ . correspondence classes of every single attribute. The second
stage computes each attribute dependency degree. The third
The accuracy of each subset approximation of X ⊆ U with phase determines the maximum dependency degree of attribute
B( X ) and ranks them based on the dependency degree.
respect to B ⊆ A can be measured by α B = , where X
B( X ) Algorithm: MDA
symbolizes the cardinality of X. Input: Data set with categorical value
Output: Nominated the most determinant
C. Dependency of Attributes attribute
Begin
The idea of the dependency attribute in the IS, given as follows.
Definition 1. Let S = (U , A, V , f ) be an IS and let D and
Step 1. Use the indiscernibility
relation on each attribute for the
C be any subsets of A. Attribute D is called be governed by equivalence class calculation.
completely on attribute C, denoted C Ÿ D , if each attribute Step 2. Define the dependency degree of
value of D is exclusively determined by value of C. attribute ai with respect to all aj,
where ij.
Step 3. Select the maximum of dependency
Definition 2 (See [13]). Let S = (U , A, V , f ) be an IS and degree of every attribute.
Step 4. Arrange the attribute in
let D and C be any subsets of A. Dependency attribute D on C in
a degree k (0 ≤ k ≤ 1) , is symbolized by C Ÿ k D . The degree k
ascending order based on the maximum of
attribute dependency degree.
is expressed End

PosC Fig. 1. The Pseudocode of MDA.


k=
U

177
2017 5th International Conference on Electrical, Electronics and Information Engineering (ICEEIE)

III. DATASET TABLE IV. SAMPLE OF DATASET

This section describes the characteristics of the used ID EE AAT 1st 2nd 3rd 4th Probability
datasets. This dataset was obtained from the Directorate of
1 B AB B B AB A 75%
Information Systems (SISFO), Telkom University. It contains 2 B B B B B A 75%
1000 instances and eight (8) categorical attributes. The attributes 3 BC C BC B BC BC 25%
that used are Student ID, Entrance Examination (EE), Academic 4 AB A AB AB AB A 100%
Aptitude Test (AAT), 1st GPA, 2nd GPA, 3rd GPA, 4th and 5 AB AB A AB AB AB 100%
Probability of graduated on time. Here, irrelevant attributes such 6 B BC B AB B BC 50%
as name, gender, student residential address, etc. have been 7 C C D C BC C 0%
removed. The occurrence of missing values might be due to 8 A A A AB A A 100%
9 B AB B AB AB A 75%
several possibilities, such as the student was on leave, the GPA 10 AB AB AB A AB AB 100%
(Grade Point Average) score is not final, the student is not
enrolled in certain semester, etc. The Probability attribute
represents the probability of graduated on time of students IV. RESULTS AND DISCUSSION
during their studies. The description of each attribute of the
dataset is shown in Table II. This research aims to identify the most determinant attribute
from the student performance dataset by creating the relevant
TABLE II. DESCRIPTION OF DATASET ATTRIBUTES attributes rank derives from the maximum-dependency of the
dataset attribute. Selecting and early knowing the most
Attribute determinant attribute, such as student’s GPA, might possibly
Description Attribute Set Value
Name
recommend the process and strategy of intervention, to conduct
ID Student ID {1,2,3,",1500} a treatment for improving student’s academic performance.
Entrance Examination
EE
score
{A, AB, B, BC, C, D, E}
TABLE V. THE RESULT OF MAXIMUM DEPENDENCY ATTRIBUTES
Academic Aptitude Test
AAT
score
{A, AB, B, BC, C, D, E}
Degree of Dependency Max
Student GPA in 3rd
1st
semester
{A, AB, B, BC, C, D, E} EE 0.0010 0.0020 0.0010 0 0 0.0020
Student GPA in 4th
2nd
semester
{A, AB, B, BC, C, D, E} AAT 0.0020 0 0.0010 0 0 0.0020
Student GPA in 5th 1st GPA 0.0020 0.0010 0.0010 0 0 0.0020
3rd
semester
{A, AB, B, BC, C, D, E}
nd
2 GPA 0.0020 0.0010 0 0.0050 0 0.0050
Student GPA in 6th
4th
semester
{A, AB, B, BC, C, D, E} rd
3 GPA 0 0.0010 0 0.0040 0 0.0040
Probability of graduated
Probability
on time
{0%,25%,50%,100%} 4th GPA 0 0.0010 0 0.0010 0 0.0020

The values of Examination Entrance, Academic Aptitude


Test, and GPA are represented by a letter, correcpondance with
its 4-scale numeric score. The actual-score conversion of
Examination Entrance, Academic Aptitude Test, and GPA
represented to a letter based on a standard that is implemented
by Telkom University is as shown in Table III.

TABLE III. GPA CONVERSION

Range of GPA GPA Conversion Category


3.501–4.000 A Excellent
3.001-3.500 AB Very Good
2.501-3.000 B Good
Fig. 2. The scatter plot of the data 2nd GPA respect to other
2.001-2.500 BC Fair
1.501-2.000 C Satisfactory MATLAB version 7.14.0.334 (R2012a) is used for the
1.001-1.500 D Passing
technique development, experiment and validation. A processor
of Intel Core i3-3217U CPU @ 1.80Ghz, with total main
0.000–1.000 E Poor
memory 4G of RAM is used to executed the techniques. The
operating system is Windows 8.1.
In Table IV, 10 out of 1500 sample of instances that are used Table IV shows that the 2nd GPA attribute is the most
as a dataset. dominant as compared to other attributes, then followed by 3rd
GPA, Entrance Examination, 1st GPA, Academic Aptitude Test
and 4th GPA attribute, respectively. Fig. 2 shows the scatter plot

178
2017 5th International Conference on Electrical, Electronics and Information Engineering (ICEEIE)

of the data which is the points inside the border line represent [4] S. K. Yadav, B. Bharadwaj, and S. Pal, “Mining Education data to predict
the positive area of the attribute 2nd GPA respect to other student’s retention: a comparative study,” arXiv preprint
arXiv:1203.2987, 2012.
attributes. The higher number of points inside the higher degree [5] E. Sutoyo, M. Mungad, S. Hamid, and T. Herawan, “An Efficient Soft
dependency implies more dominant than other attributes. Set-Based Approach for Conflict Analysis,” PloS one, vol. 11, no. 2, p.
e0148837, 2016.
V. CONCLUSION [6] S. Zainudin, D. S. Jasim, and A. A. Bakar, “Comparative Analysis of
Data Mining Techniques for Malaysian Rainfall Prediction,”
This paper presents the use of Maximum Dependency International Journal on Advanced Science, Engineering and
Attributes (MDA), derives from rough set theory as the most Information Technology, vol. 6, no. 6, pp. 1148–1153, 2016.
determinant attribute selection by ranking the relevant attribute.. [7] D. W. Jacob, M. F. M. Fudzee, M. A. Salamat, R. R. Saedudin, I. T. R.
We elaborated the technique using the dataset obtained from the Yanto, and T. Herawan, “An Application of Rough Set Theory for
Directorate of Information Systems (SISFO), Telkom Clustering Performance Expectancy of Indonesian e-Government
University. The dataset contains 1000 instances and eight (8) Dataset,” in International Conference on Soft Computing and Data
Mining, 2016, pp. 638–646.
categorical attributes. The attributes that have been used are [8] D. W. Jacob, M. F. M. Fudzee, M. A. Salamat, R. Saedudin, Z. Abdullah,
Student ID, Entrance Examination, Academic Aptitude Test, 1st and T. Herawan, “Mining Significant Association Rules from on
GPA, 2nd GPA, 3rd GPA, 4th GPA, and Probability. The result Information and System Quality of Indonesian E-Government Dataset,”
showed that Maximum Dependency Attributes can be used to in International Conference on Soft Computing and Data Mining, 2016,
select the most determinant attribute of student performance pp. 608–618.
[9] M.-L. Antonie, O. R. Zaiane, and A. Coman, “Application of data mining
dataset. The most determinant attribute is 2nd GPA, followed by techniques for medical image classification,” in Proceedings of the
3rd GPA, Entrance Examination, 1st GPA, Academic Aptitude Second International Conference on Multimedia Data Mining, 2001, pp.
Test and 4th GPA attribute, each. Therefore, our future work will 94–101.
be more focused on how to design strategic programs and [10] E. Sutoyo, I. T. R. Yanto, Y. Saadi, H. Chiroma, S. Hamid, and T.
conduct for well-planned improvement on student academic Herawan, “A Framework for Clustering of Web Users Transaction based
on Soft Set Theory,” 2015.
achievement. [11] R. R. Saedudin, S. B. Kasim, H. Mahdin, and M. A. Hasibuan, “Soft Set
Approach for Clustering Graduated Dataset,” in International
REFERENCES Conference on Soft Computing and Data Mining, 2016, pp. 631–637.
[1] Z. Ibrahim and D. Rusli, “Predicting students’ academic performance: [12] T. Herawan, I. T. R. Yanto, J. M. Zain, Q. Hongwu, Z. Abdullah, and
comparing artificial neural network, decision tree and linear regression,” others, “A soft set approach for clustering student assessment datasets,”
in 21st Annual SAS Malaysia Forum, 5th September, 2007. Journal of Computational and Theoretical Nanoscience, vol. 12, no. 12,
[2] C. Márquez-Vera, A. Cano, C. Romero, and S. Ventura, “Predicting pp. 5928–5939, 2015.
student failure at school using genetic programming and different data [13] T. Herawan, M. M. Deris, and J. H. Abawajy, “A rough set approach for
mining approaches with high dimensional and imbalanced data,” Applied selecting clustering attribute,” Knowledge-Based Systems, vol. 23, no. 3,
intelligence, vol. 38, no. 3, pp. 315–330, 2013. pp. 220–231, 2010.
[3] I. T. R. Yanto, P. Vitasari, T. Herawan, and M. M. Deris, “Applying [14] Z. Pawlak, “Rough sets,” International Journal of Computer &
variable precision rough set model for clustering student suffering Information Sciences, vol. 11, no. 5, pp. 341–356, 1982.
study’s anxiety,” Expert Systems with Applications, vol. 39, no. 1, pp.
452–459, 2012.

179

View publication stats

You might also like